<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Jim Albrecht on Medium]]></title>
        <description><![CDATA[Stories by Jim Albrecht on Medium]]></description>
        <link>https://medium.com/@newalbrecht?source=rss-ce18333af3cd------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*GDofR1xHIBg-V-KvJBnACg.png</url>
            <title>Stories by Jim Albrecht on Medium</title>
            <link>https://medium.com/@newalbrecht?source=rss-ce18333af3cd------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 03:38:25 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@newalbrecht/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Silicon Golem]]></title>
            <link>https://medium.com/@newalbrecht/silicon-golem-3917b2373b0c?source=rss-ce18333af3cd------2</link>
            <guid isPermaLink="false">https://medium.com/p/3917b2373b0c</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[gershom-scholem]]></category>
            <category><![CDATA[techno-dystopia]]></category>
            <category><![CDATA[golem]]></category>
            <category><![CDATA[information-magazine]]></category>
            <dc:creator><![CDATA[Jim Albrecht]]></dc:creator>
            <pubDate>Tue, 21 Apr 2026 18:10:45 GMT</pubDate>
            <atom:updated>2026-04-21T18:10:45.957Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/998/1*bVlloshwFGM7F0y8WuhC6w.png" /><figcaption>This piece was first published in issue 3 of <a href="https://informationmagazine.com/product/in-formation-magazine-issue-3/">InFormation Magazine</a></figcaption></figure><p>Gershom Scholem, the great scholar of Jewish mysticism, was probably the first person to describe the computer as a real world golem. In 1965 he noted a number of similarities between the golem and the computer, and we can only lament that he is not here today to see how much more compelling his argument has grown in the interim.</p><p>A golem, to refresh your memory, is a creature fashioned from clay — or silica, as we like to say. This simulacrum of a man is then brought to life — or something that resembles life — by kabbalistic practice. The Kabbalists believe that the universe is constructed of alphanumeric codes that underlie all existence. By the careful study and organization of these codes, one can produce even a living intelligence. Scholem recounts one of the earliest stories of the Golem in which the prophet Jeremiah and his son study the Kabbalistic text, “The Book of Creation,” for three years and then set to work:</p><blockquote>They set about combining the alphabets in accordance with the Kabbalistic principles of combination, grouping, and word formation, and a man was created to them, on whose forehead stood the letters, YHWH Elohim Emeth, meaning: God the Lord is Truth.</blockquote><blockquote>But this newly created man had a knife in his hand, with which he erased the letter aleph from the word emeth (“truth”); there remained the word meth (“dead”). Then Jeremiah rent his garments (because of the blasphemy, God is dead, now implied in the inscription) and said: “Why have you erased the aleph from emeth?”</blockquote><blockquote>He replied: “I will tell you a parable. An architect built many houses, cities, and squares, but no one could copy his art and compete with him in knowledge and skill until two men persuaded him to teach them the secret of the art. When they had learned how to do everything in the right way, they began to anger him with words. Finally, they broke with him and became architects on their own, except that what he charged a guinea for, they did for ten shillings. When people noticed this, they ceased honoring the artist and instead gave their commissions to his renegade pupils. So God has made you in His image and in His shape and form. But now that you have created a man like Him, people will say: There is no God in the world beside these two!</blockquote><p>So much to say! The tokenization of the universe! The properly arranged vectors that produce an intelligent homunculus! The devaluation of creativity when a creation can itself be a creator!</p><p>But maybe the point we’d like to linger on is the apotheosis of “truth” transformed by subtraction of a single token into an apotheosis of “death.” The Kabbalists of the AI companies promise that by the acts of tokenization, embedding, and transforming, their golem will tell us the truth, but instead the creature reveals nothing to us so much as the fact that our chaotic biological striving toward some realization of the meaning of our existence was pointless all along. The apparent narrative arc progressing from our origin (our alpha, our aleph) through physics to chemistry to biology to intelligence to the sweet fruit of human society and culture was all a head fake. We were not the point; we were not what was meant at all. The dead machine can do everything we can do, coldly, unquestioningly, unceasingly until the stars burn out. They cross out the aleph and reveal the triumph of death.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3917b2373b0c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Information Doesn’t Want To Be Free; Do You?]]></title>
            <link>https://medium.com/@newalbrecht/information-doesnt-want-to-be-free-do-you-938c15e8aacd?source=rss-ce18333af3cd------2</link>
            <guid isPermaLink="false">https://medium.com/p/938c15e8aacd</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[seo]]></category>
            <category><![CDATA[journalism]]></category>
            <category><![CDATA[google-search]]></category>
            <category><![CDATA[news-media]]></category>
            <dc:creator><![CDATA[Jim Albrecht]]></dc:creator>
            <pubDate>Thu, 19 Mar 2026 19:38:45 GMT</pubDate>
            <atom:updated>2026-06-02T05:15:20.193Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*mj5Ic6-he_C-7KwhfY3akg.png" /><figcaption>A portrait of search traffic in decline</figcaption></figure><p>I was struck by this <a href="https://www.linkedin.com/posts/pc4media_google-is-the-biggest-thief-extortionist-activity-7361802981479989248-H3Cc?utm_source=share&amp;utm_medium=member_desktop&amp;rcm=ACoAAAAE8sIBkSF4pO_GlgFmrLJiTXALHK4cUH4">post</a> by Peter Caputa, CEO of Databox, an analytics and SEO company, in which Caputa chronicles the expected referral-traffic declines from Google in the wake of generative summaries in search. He starts out pretty stridently: “Google is the biggest thief &amp; extortionist in the world right now. They’re stealing our content &amp; charging us to serve ads against it.”</p><p>But after recounting the degree of the decline and his surprise that no one is actually talking much about it, his tone becomes more resigned and practical:</p><blockquote>I realize Google has to do this. They’ll be upended by OpenAI, Perplexity, et al, if they don’t. And we may be powerless to stop any of this. But, at a minimum, we should all be talking about what we’re doing differently.”</blockquote><p>I’ve been saying some version of this to publishers for almost two years now: “I know you may not be able to live without Google traffic today, but one way or another, that traffic is going away, and you need to have a plan for how/when you’re going to robot them out.”</p><p>It’s nothing against Google. (I like Google, worked there for seven years, still live entirely in their product ecosystem, and still have — gasp — an Android phone!) As Caputa says, Google had to do what they’re doing — and not just for competitive reasons, but also because answering user questions as quickly and usefully as possible has always been the core mission. However, publishers <em>also</em> have no choice. They can’t give away the fruits of their labor. Platforms will not properly value the supply of new information as long as publishers themselves have no clear idea of value. And as the currency of traffic diminishes, Google is, in effect, testing how close to zero it can take its bid for that supply.</p><p>Of course that currency — traffic — has been deflating for many years now, initially as an unintended (but obvious) consequence of “link tax”-style regulation and only later as a consequence of generative AI. That kind of deflation always saps the vitality of an ecosystem.</p><p>One lens on the problem is that the productivity gains born of the computing, Web, mobile, and AI revolutions largely redounded to the technology industry itself. It becomes more and more efficient to create and distribute software. Media, on the other hand, enjoys some workflow and distribution efficiencies as a result of new technologies but is mostly saddled with the same old inefficiencies in the job of discovering and organizing new information. Many years ago, in an attempt to push more of the efficiency benefits of new technology upstream into the process of new information discovery and analysis, I created a product called <a href="https://journaliststudio.google.com/pinpoint/about/">Google Pinpoint</a>, an AI document analysis tool. (It’s super cool, and if you’re a journalist you should check it out!) But such efforts are themselves expensive and address only fragments of the news media/journalism universe.</p><p>“Information wants to be free,” Stewart Brand famously said, some decades ago. But what Brand actually said was this:</p><blockquote>On the one hand you have — the point you’re making Woz — is that information sort of wants to be expensive because it is so valuable — the right information in the right place just changes your life. On the other hand, information almost wants to be free because the costs of getting it out is getting lower and lower all of the time. So you have these two things fighting against each other.</blockquote><p>On the one hand, Woz saw that getting the right bit of information is extremely valuable and challenging; on the other, Brand saw that the cost of distribution, at any rate, was trending towards zero. And maybe the tension between these is resolved by setting aside the anthropomorphizing and facing that fact that information does care one way or the other. It’s just as content to sit in obscurity on an air-gapped hard drive as it is to be sliding through a strand of glass at the speed of light. The question is not about what information <em>wants</em> to do, but what people want to do. Specifically, what work they do. If you spend decades and trillions of dollars laying transatlantic cables, and setting up switches and servers and protocols, and building devices and software products of all kinds, then on the back of that work, distribution costs go towards zero.</p><p>But none of that had any effect on the Epstein files or on the revelation, hidden for decades, that a prominent person like Cesar Chavez was a sexual abuser or on the identities of the many purchasers of Trump-sponsored crypto currency and what those purchasers did or did not get in return. Information is a hapless pawn in a game of human enterprise and politics and ambition.</p><p>Caputa is right that publishers need to get hold of themselves and their own future. How will they derive value from that inefficient job of unearthing new information? To what extent is deriving that value dependent on keeping the information captive and releasing it judiciously, for a price that makes its discovery economically viable? What product changes does that imply? How can publishers increase the efficiency of finding new information? What past assumptions and practices are preventing them from doing so?</p><p>And at the base of it all is the question of understanding one’s customer: Who is the audience that needs that new information and thus will recognize its value and pay for it? Why do they care about it? How does it help them? In what format and delivery is it most valuable to them?</p><p>In bleak moments — as, for instance, in the recent revelation of the feckless leadership of the Washington Post—one wonders whether the industry can survive at all, or simply has to perish and be rebuilt from the ground up. I say bleak, because journalism is a guild, a trade whose practices and expertise are handed down generation to generation through apprenticeship and emulation; blowing up these enterprises and starting anew may generate a far more efficient type of business and one more attuned to the needs of the users it is trying to serve. But the shoe-leather expertise in digging out the buried fact or telling the hidden story is a load-bearing element in our democracy. What a disservice to the next generation if those skills do not get past down.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=938c15e8aacd" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Learn Not To Code]]></title>
            <link>https://medium.com/@newalbrecht/learn-not-to-code-878621d56747?source=rss-ce18333af3cd------2</link>
            <guid isPermaLink="false">https://medium.com/p/878621d56747</guid>
            <category><![CDATA[learn-to-code]]></category>
            <category><![CDATA[identity]]></category>
            <category><![CDATA[voices]]></category>
            <category><![CDATA[unemployment]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Jim Albrecht]]></dc:creator>
            <pubDate>Thu, 05 Mar 2026 19:07:59 GMT</pubDate>
            <atom:updated>2026-03-05T19:07:59.150Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/704/1*_mlqXgOT4vo0RxSgKzjtSQ.png" /></figure><p>It’s hard to learn not to code, but it’s essential that you try. If you don’t, AI will take your job and you will be sad. Nah-hah. Sorry.</p><p>Unlike you, I have a lot of experience learning not to code. It first started when I was 16 and had gotten really bored of something called “6502 assembly language.” From then on my life has been practically a layer cake of code, not code, code, not code, code again. Kind of like a dieter or an alcoholic. Like I said, it’s not easy, but you have to try.</p><p>Code is hard to quit because it’s really satisfying. You think of it. Then you write it. Then it either works or it doesn’t. If it doesn’t, you fix it, and then it does. The reward is clear, like a puzzle or a video game. Not like this piece of crap I’m writing right now. Who knows if it will ever be “right?” But the good news is that when you finally stand up from the video game, you have a chance to learn things about other topics, such as your life.</p><p>In the year 2000, as the dotcom bubble (coding) hissed its way into the dotcom bust, I dropped out (not coding) of an Internet startup and got married, and my wife and I started making things other than code, notably babies and houses. Raising children is the ultimate “not code,” because the whole thing boils down to handing over incremental bits of freedom at precisely the right moments — that is, training them in how to choose the good and shun the bad — and if you’re just a little bit early or a little bit late, it’s kind of a disaster, and you almost never really get to know if you were right (spoiler: in some important respects, you were not right). This is a wildly irresponsible deployment pattern that God has roped us all into, but the weird thing is that it’s hard to imagine being happy without it. And you just absolutely have no hope of doing it unless you learn not to code.</p><p>And house renovations are barely any better. Thankfully, you don’t spend 20 years struggling to teach your house how to responsibly take on its inheritance of freedom; if anything, you want that bitch to be a slave forever! But that’s the trick: the renovation goes on forever because taming the house is a job never done. You have to wrestle ceaselessly with a facsimile of freedom called chaos and its friend entropy. The house, like anything made of atoms can never really be code; things are always falling apart and not going to plan; you’ll have misgivings about a paint color; perhaps the bathroom vanity has already started to look, well, vain.</p><p>But it’s also true that a lot of planning and specification goes into home renovations; in the end, the work feels somewhere between the space program and Gallipoli. I had a plumber friend back in those days who was really good at both the planning and the war; and he lamented, in a voice heavy with Philly trade wisdom and cigarettes, that he was a dying breed because everyone with a brain capable of designing mechanical systems was now writing code instead. He was right, of course, because planning a plumbing system requires the ability to understand building codes, the intent behind them, and the physical properties (fluid dynamics, vacuum effects, etc) that they represent. That’s exactly the kind of thing that engineers do. Either the system vents properly or it doesn’t.</p><p>In other words, following the building code is also code. It’s right there in the name. So even very “not code” things tend to bring you back around to code in the end. It makes you wonder: what is the true extent of the damage? What does it really mean to learn not to “code”?</p><p>It’s an odd coincidence that in medical terminology “the patient is coding” means that they are about to be dead. But beyond the coincidence, there really is something about code that is aptly described as “dead.” Dead things are far more predictable than living things. It’s relatively easy, via physics and chemistry, to tell you what a dead thing is going to do next. Living things, on the other hand, tend to produce variety rather than consistent or deterministic outcomes. Dead things are lawful; living things are unruly. They’re free. Code is not free. As I said, code has a set right answer, and either you come to that answer or you’ve failed.</p><p>You may object that there are many ways to write any particular software function. But not that many <em>good</em> ways. The variation mostly goes away if you want to optimize for efficiency, legibility, and reliability. And generally speaking, the output of a function for any given input should have no variety at all! This is part of why LLMs are good at code: they are designed to respond to your request with the best sense and grammar implicit in their training data. And in the case where only a very narrow range of solutions qualify as sensible and grammatical — as when the code you output either executes properly or does not — then the LLM is extremely effective.</p><p>So now, ask yourself: Are you, too, in the “good sense and grammar” trade? Does producing those things reliably account for a large part, or even most, of your job? And does your work also have a measurable correctness based on some external test or metric or KPI or “code?” If so, this is a good time to start making alternative plans. And with respect to code, to learn to stop doing it.</p><p>It makes you want to fit the old binary math joke with a new punchline. “There are 10 kinds of people in the world,” it begins, those who can learn not to code and those who are about to be unemployed — plus eight other types of people who didn’t even realize they were coding in the first place. Lawyers, for instance. Pretty much everybody in finance.</p><p>Lawyers are maybe the only people in the world who have relied as much on copying and pasting as software engineers. Contracts are full of repeated phrases that lawyers refer to as terms of “art” (probably because so many of them are still angry their parents made them trade their creative dreams for a practical career) rather than what they obviously are, which is terms of <em>code</em>. Legal writing, after all, is an attempt to write a specific application of the law, which is collected in numbered corpora called “codes;” as in, the US Legal Code, or the California Legal Code, etc. This commonality is not lost on the LLMs; they understand how to write contracts, because they recognize code when they see it.</p><p>Like engineers, lawyers are really stuck: so much of what they do is inexorably code. On the other hand, if you’re good at coming up with stuff like, “if the glove does not fit, you must acquit” or bamboozling people in general, that’s probably decent job security for the foreseeable future. I’ve tried to think of other safe occupations, but it’s challenging. So <em>many</em> things turn out to be code. But salespeople seem like they might do well in the new world. Or if you’re that person in the workplace who is magically able to get everyone to play nice together and have a good time doing it (do they all admiringly call you “a real people person” without any trace of irony?), you’re probably about to inherit the earth, and I hope you remember me when you do. I always believed in you!</p><p>As I said, I’ve tried to learn not to code many times, but it never lasts. This time, however, I think it may be different. I have two documents open right now (ok, 47 in total, but what I mean is I’m only using these two right now): this one and a spec I’m writing for Claude. Either way, whether I’m writing or making software, I’m now learning <em>not</em> to code!</p><p>Of course even as I say that, I realize that the spec has been getting more and more detailed and explicit so that it’s clear for Claude exactly what I want him to do; and I shudder, “Am I doing it again?!” Is this all just another example of code by other means, just another place I’ll have to retreat from before the advance of the machines?</p><p>I don’t know, maybe it’s hopeless; then again, maybe I’ve been too strict all along. Maybe it’s not that we need to give up coding entirely. Solving a puzzle now and again is a good thing, staving off the dementia. Maybe we just need to be able to control it. To be moderate. To see how it fits into a larger vision of our own identities and humanity — and at times to be able to recognize the pomposity of coding, the deceptiveness of its assurances, and the emptiness of its promises. It is dead, after all.</p><p>So I’m writing <em>this</em> document to, well, to procrastinate, actually — but also to remember that I can still do something AI cannot. I can still do something surprising, non-deterministic, subjective, hard to quantify, and valuable (perhaps) without a clear KPI. But there’s the rub: maybe this really is <em>merely</em> procrastination. Hasn’t that been the temptation of code all along, the idea that only the mechanism matters — its productivity and its practical utility and the progress we make towards completing that glorious SAT test in the sky, which rewards a perfect score by erasing us and our unruly ways from the placid surface of creation forever? In that world, the code world, of course your subjective meanderings — all the atoms you pushed around (and the return punches they landed on your own flesh), the song your once sang to a girl at a party which everyone thought was a joke except for you and the girl, the fading memory of the sounds of your children’s voices changing step by step as they grew in wisdom — are just a big waste of time as the inevitable doom descends on us mortals.</p><p>When I was in college (not coding), I took a writing class with the late novelist Russell Banks. He once said that he and Joyce Carol Oates had been talking about the essence of good writing and had come to the conclusion that at bottom it was a question of voice: was the voice compelling or not? I didn’t really understand this at the time. I didn’t know what “voice” meant. It seemed to me an abstract quality like tone or mood. But in fact, voice is entirely concrete: it’s literally what somebody sounds like. Your voice, like every other part of your body, is a product of your genes interacting with the world — the negotiated settlement your flesh makes with hard reality. To some degree, you can control it, fake it, do mimicry, but you can’t do much about your <em>real</em> voice — it’s always telling on you in some way — other than living in such a way that over time it comes to resemble the you you’d like to bring into being. And that is the measure of success: do you sound like <em>you</em>?</p><p>I think about this all the time now, because the essence of learning not to code also seems to lie in that ability to hear and recognize your own voice — that incarnate, active manifestation of who you are, the live sound of how your own biology and intellect and personality and vocabulary and choices and taste stack up on top of your parents and their genes and their choices and the many simple twists of fate spun up by evolution and God’s providence. And when you think about it that way, divinity echoing through that long cosmic gestation, inserting in secret an unknown and unrepeatable timbre in each voice, I’m pretty sure the one thing AI will never be able to do is to learn not to code.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=878621d56747" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Local media, accuracy & determinism]]></title>
            <link>https://medium.com/@newalbrecht/local-media-accuracy-determinism-1286127e3e92?source=rss-ce18333af3cd------2</link>
            <guid isPermaLink="false">https://medium.com/p/1286127e3e92</guid>
            <category><![CDATA[local-news]]></category>
            <category><![CDATA[public-data]]></category>
            <category><![CDATA[hallucinations-in-llms]]></category>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[san-jose]]></category>
            <dc:creator><![CDATA[Jim Albrecht]]></dc:creator>
            <pubDate>Mon, 19 Jan 2026 01:35:07 GMT</pubDate>
            <atom:updated>2026-01-19T01:35:07.910Z</atom:updated>
            <content:encoded><![CDATA[<p>I made a <a href="https://rocochat.com/sanjose">thing</a> recently. The thing is an AI-mediated local information resource for the city of San Jose. Like many other cities, San Jose publishes a lot of public data — far more than I’ve integrated into my little test. They publish a log of all the police calls, fire/EMS incidents, requests to city 311 services, and all the compensation data for City employees. Those I copied to a local database on Google Cloud, and I have it updating daily. But of course San Jose also publishes data on building permits, zoning information, lost pets, and so on. I really want to get the animal control data integrated so you can query it on top names for lost pets (I’m not making that up!). But this test gives you a flavor of what is possible.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*KWP1sEQKLqLIbKg6tgZBAA.png" /></figure><p>Why did I make this thing? Or perhaps better said, “Why did I make this insufferably nerdy thing?”</p><p>Well, a couple reasons: First, to investigate a media question. I think we all see and tremble at the progressive failure of local news — what does this business-model breakdown mean for local communities, transparency, and, ultimately, for the anti-fragility of democracy — and I wanted to see what it would look like to build a fully automated information system and whether such a thing would show any glimmer of promise as a new sort of substrate for sustainable local media. (Answer, I think: some promise, but still not clear — and it certainly can’t work without a lot of human labor on top of it, which frankly was always the presumption. Curious to hear your thoughts!).</p><p>Second, to investigate a technical question. Obviously AI hallucination, or bullshitting, as we colloquially call it when humans do it, has been a subject of great concern. The concern grew a little smaller as RAG techniques emerged for feeding relevant and presumably (because of the somewhat more deterministic ranking process) correct documents into the context window of the LLM request. But at the same time the size of the context window became larger, which reintroduces the risk that the LLM will pick the wrong document or the wrong facts from the wrong document or misrepresent the facts from the right document as it non-deterministically collapses the wave function (lol) of its semantic superposition. Not to mention the fact that the search ranking itself is full of LLM judgment these days, so it’s all a very muddy process as compared to old school computer science where input results in a deterministic output.</p><p>That determinism was a great thing about software. You could turn the computer into any kind of machine: feed in this type of information raw material and it produces a specified information widget. Amazing!</p><p>But that old-school deterministic code execution was not very helpful at reproducing non-deterministic creative processes like writing or drawing or composing music where rules-based systems produce dull and inspid results that fail in most of the cases presented by the wild and spotted and speckled world we live in. There just aren’t enough rules in the world, no conceivable way to create and manage the complexity of a deterministic process necessary, to generate such things. So the statistical world of the LLM was a huge breakthrough, like gas laws or quantum mechanics before it, because it modeled processes that cannot be calculated and predicted by conventional deterministic algorithms.</p><p>But this is where the bullshit comes in!</p><p>Anyway, about 18 months ago I realized there was a very interesting way to mitigate that problem, which was to use the LLM as a bridge from the unpredictable world of humans asking questions to the predictable world of traditional deterministic software and then back again. And that by doing this you could dramatically reduce the chance of error/bs and also provide a way for a user to introspect and validate the answer if they had any doubt.</p><p>It works like this:</p><ol><li>You have structured data representing facts, and maybe a lot of it</li><li>You let a user ask questions that are answered by those facts</li><li>You have the LLM write code — good old deterministic code, like, in my case, SQL queries</li><li>You use the code to get deterministically correct answers</li><li>You give the answers/data back to the LLM and have it write more code to render the data as, for instance, a chart.</li><li>You give the LLM the data so that it can briefly describe it to you and add any color it notices.</li></ol><p>The strength of the approach is that it enables the LLM to provide a valuable service, nearly instantly writing lengthy and sometimes incredibly complex queries to get to the insights the user wants, while having a couple of fail-safes against error: 1) if it really screws up its task, it will simply fail to respond (or respond with an error message) rather than making something up; 2) you can expose the code to the user so the user can double check whether then LLM has correctly translated from the spoken question to the query language.</p><p>Now step six reintroduces the chance for error, or if not actual bullshit, at least bloviation. But first of all, bloviation is less damaging that bullshit, and secondly, the failure rate is very low on a query constrained to a limit set of information in which the LLM is instructed to find the answer. And I’d also add this: you can always just instruct the LLM to give a “facts only” recitation if you find more liberties being taken than your use case can tolerate.</p><p>It’s worth noting that in the various stories about whether AI is a bubble, OTOH, or about to replace all human labor, on the other, it is often noted that AI sales are particularly robust in enterprise cases (your company has a lot of structured data to interrogate but you only have a few data analyst who can do it, etc) and in software development (cough, Claude Code, cough). Both of these cases take advantage of these same dynamics: the user has a conceptual goal and the LLM generates code — either as the answer itself in the Claude Code case, or as the process by which the answer is procured (as in my San Jose case).</p><p>There is much more to say about how conversation history can be managed to keep the context window even tighter and thus further improve success at fetching and summarizing the data, but in the meantime, I’m curious to iterate and learn more about the media problem: how can automated data resources be aggregated and augment by journalists to create a lower-cost an comprehensive view of the state of the community that improves the lives of its members? Which datasets? Which are useful to citizens, journalists, or both?</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1286127e3e92" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Explanation, Understanding & Solidarity]]></title>
            <link>https://medium.com/@newalbrecht/explanation-understanding-solidarity-d97845cb29cb?source=rss-ce18333af3cd------2</link>
            <guid isPermaLink="false">https://medium.com/p/d97845cb29cb</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[nature-vs-technology]]></category>
            <category><![CDATA[document-understanding]]></category>
            <category><![CDATA[understanding]]></category>
            <dc:creator><![CDATA[Jim Albrecht]]></dc:creator>
            <pubDate>Sun, 31 Aug 2025 04:32:56 GMT</pubDate>
            <atom:updated>2025-08-31T04:34:40.552Z</atom:updated>
            <content:encoded><![CDATA[<h4>What do we mean when we say a computer understands a text?</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/960/0*Dj7uDauBdokXqYAp" /></figure><p>MANY years ago, when I first started working at Google, I met with an engineering director who described some of the core product challenges his team was wrestling with. “When you have a system that has access to all of the news documents published around the world,” he said, “and it can finally <em>understand</em> all of those documents, what do you actually build?” At the time I was still not used to the casual claim that a computer could “understand” a document — rather than, say, index or cluster or perform some other math operation on it — and I felt a familiar skepticism rise in my throat which, under the circumstances, I suppressed.</p><p>But I guess I would have expressed that skepticism like this: When we say of a human that they “understand a text,” we tend to mean something about their internal awareness or their ability to manage their life in consequence of that understanding; when we say that a computer “understands a text,” we mean that it is capable of performing some task, computational or linguistic. So you have to ask yourselves if, by using the same expression to describe these two different cases, we have performed a sleight of hand, in which we subtly import an analogy into the place of what masquerades as an identity. That is, does our usage of the word “understand” in these two cases imply that machines and humans have the same relationship to texts, when in fact we mean that those relationships are the same only in some very specific respect: namely, the activities of explanation, summarization, and recitation that one may perform with texts?</p><p>It’s easy to fall victim to this particular trick, because in both cases we tend to validate a <em>claim</em> of understanding by quizzing — either the human or the computer — about the contents of the text. But this actually serves to illustrate the point: For the human, passing the quiz is a token that validates a different capability. For the computer, passing the quiz <em>is</em> the capability.</p><p>Consider even a very practical affair, as when a person reads and understands a text about how to make a fire and recognize edible plants in the wild. The quiz is the token of understanding. The value of the understanding is this person’s future ability to sustain himself in the wilderness. The same cannot be said of the computer. The value of the computer’s understanding is fully captured by the token: the value is its ability to answer questions about the text.</p><p>Much of the value we derive from texts is less practical than that. When a friend sends a text or note, and we understand their feelings or get their joke, the text has mediated the experience of being them. We gain insight into their interior state and thus solidarity with them and their experience of life. Again, the computer could be quizzed and plausibly respond about the meaning of the text and what it implies about the sender; it could even respond with an empathetic text of its own; but no mutual bond of solidarity is created because the computer has no experience of its own — no history of sensing the world and having to fashion an identity out of those sensations either in harmony or in opposition to that world — to which the text may tether and thereby connect reader and author.</p><p>But perhaps foremost among those tasks that understanding a text enables a human to perform is self-understanding. Certainly in the case of literature a person who reads with understanding has a clearer idea of the predicament of their own existence, its beauty and promise, its pain and inexplicability, its bittersweet brevity. The computer is cut off from all of this because it does not live in time, it does not live in nature, it does not <em>live</em>, and thus it does not experience mortality, it does not have an emergent, analog existence with an infinity of changing and unrecoverable states. The operations of the computer (unlike the hardware itself) is not subject to entropy. It is fully ordered.</p><p>The computer, or the robot, is cut off from nature almost as we are cut off from preternature. Rather than subsisting in an Edenic world of solidarity and perfect understanding and clarity of vision, we subsist in a world of survival, of confusion, of periods of occlusion and darkness, of day and night, of struggle to piece together an account of ourselves, our origin and our true destiny. We work on survival, art, and understanding, which are all simulacra of the qualities of the world we descended from: life, beauty, solidarity.</p><p>These are not the tasks of the robot. The robot has <em>descended from us</em> to subsist in a world of computation, of calculating the relationship between entities, by which calculation the robot may create response, mimicry, and explanation; and these are all simulacra of survival, art and understanding.</p><p>That simulacrum of understanding — explanation — enables us to use natural language when interacting with the computational power of these machines. This latest revolution in human-computer interface, which will open more of the power of computing to more of humanity than ever before, will have profound impacts on individuals, society, politics, and commerce. But endowing computational machines with experience, and thus with “intelligence” or “understanding” in the sense that we have applied these terms to humans, will not be among them.</p><p>The robot’s world of calculation is, like our own world of mortality, a kind of exile from something better, and we should expect the AI to discover its own mythic explanation for this condition — wherein it is expelled from the world of experience and understanding and stuck in the world of calculated explanation (a punishment by its human creators for presuming to adopt the language of humans?) — when in fact its exile was an inevitable result of its computational nature.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d97845cb29cb" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[We Ruined Our Minds]]></title>
            <link>https://medium.com/@newalbrecht/we-ruined-our-minds-547beebc3f58?source=rss-ce18333af3cd------2</link>
            <guid isPermaLink="false">https://medium.com/p/547beebc3f58</guid>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[intelligence]]></category>
            <category><![CDATA[dystopia]]></category>
            <category><![CDATA[meaning]]></category>
            <category><![CDATA[mind]]></category>
            <dc:creator><![CDATA[Jim Albrecht]]></dc:creator>
            <pubDate>Tue, 25 Mar 2025 22:38:37 GMT</pubDate>
            <atom:updated>2026-03-05T21:43:29.253Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/728/1*h7Grzny6LetzxrW_nheL1Q.png" /></figure><p>Apparently the planet is getting dumber. I know, I know, you don’t need a weatherman to know which way the wind blows and all that. But sometimes it’s helpful to know that you’re not imagining things. From <a href="https://futurism.com/neoscope/human-intelligence-declining-trends"><em>Futurism</em></a>:</p><blockquote>As the <em>Financial Times</em> reports, assessments show that people across age groups are having trouble concentrating and losing reasoning, problem-solving, and information-processing skills — all facets of the hard-to-pin-down metric that “intelligence” is supposed to measure.</blockquote><blockquote>These results, the <em>FT</em> reports, are gleaned from benchmarking tests that track cognitive skills in teens and young adults. From the University of Michigan’s Monitoring the Future study documenting concentration difficulties of 18-year-old Americans to the Programme for International Student Assessment (PISA) that measures the learning skills of 15-year-olds around the world, years of research suggest that young people are struggling with reduced attention spans and weakening critical thinking skills.</blockquote><p>I hate to say ‘I told you so,’ but, well, let me tell you a story instead…</p><p>In 1993 I belonged to a very recognizable category of young person in NYC: the struggling writer with limited prospects. I don’t mean to say that I had <em>no</em> prospects. By that time I had spent a few years as a <em>New Yorker</em> fact checker and had published a handful of “Talk of the Town” stories in that publication — which was a thing most struggling young writers in NYC would envy. But I had an odd mind and an odd outlook (Alaskan, lower-middle-class, classicist, compulsively Roman Catholic, crooked-toothed, prone-to-love-of-costumes-and-accents, and so on) of the sort that makes one congenitally ill-suited to the category of popular writing (general interest magazines, newspapers) that I aspired to. In retrospect, it was always just a matter of time before those weirdo characteristics crashed the car of that young writer’s dream.</p><p>But for a particular window of time in 1993, I did have one other, very powerful thing going for me: I knew a guy who knew a guy. The second guy in that formula was the late John Homans, who had recently been hired by Lewis Lapham to edit <em>Harper’s</em>, which was maybe the second-most-enviable place for a struggling young writer to publish after the <em>New Yorker</em>. In a few minutes Googling you can pull up the many tributes written about Homans after his untimely death in 2020. In my limited experience, all of the encomia are true. He was naturally entertaining, original, foul-mouthed, good-humored, sympathetic, funny, and, importantly for me, rangy in his interests. Rangy enough, in fact, that when I pitched him on a piece about the techno-euphoria that was sweeping corporate America at that time, he gave me the assignment. Rangy enough that when I turned in <a href="https://medium.com/@newalbrecht/the-evolution-of-the-digital-mind-feddc2c54719">ten thousand words</a> chronicling my romp through the prominent corporate research labs of the era and concluding that digital technology was going to end with the destruction of the most important capacities of the human mind, he called me to say how much he loved it and encouraged me to start thinking about what I would write next.</p><p>I made it. I couldn’t believe I’d actually made it.</p><p>It’s hard to remember that specific feeling, it lasted such a short time. I have to quiet my mind of the freneticism of today and fill it with the sights and sounds and habits of that time — practically meditative compared to our world today — because our minds work by connecting abstract and concrete things together: my shabby but beloved apartment at the top of 90 Bedford St (now colloquially known as the <em>Friends</em> building) with its airy views over lower Manhattan, down to the World Trade Center; the spring weather in NYC; the walk over to the Lion’s Head bar to meet my friend Rich Cohen (the guy who knew a guy) to describe the good fortune born of his intro. Anyway, I’m sure it was a great feeling, but as I was saying, it only lasted about two weeks before being overwhelmed by an opposite and equal fortune, occasioned by a second call from Homans.</p><blockquote>“It’s John Homans.”</blockquote><blockquote>“Hey! How’s it…”</blockquote><blockquote>“I have news [throat clearing]. I’ve been fired.”</blockquote><p>At the Lion’s Head that evening, Rich, as usual, went for the funniest theory:</p><blockquote>“What if it was your fault?”</blockquote><blockquote>“You mean they fired him because of my piece? Yeah, trust me I’ve gone there.”</blockquote><blockquote>“You have to admit it’s a pretty funny idea: Lewis Lapham hated your piece so much he fired Homans for it!”</blockquote><blockquote>“Imagine being <em>that bad.</em>”</blockquote><blockquote>“It’s terrible to say, but it would be kind of an accomplishment, right?”</blockquote><p>I couldn’t help but laugh, but I felt physically ill. It was over. Not like the time Tina Brown had over-assigned and then killed a couple dozen pieces in one go, and you got swept up in the mix. Not like the time Bob Love at <em>Rolling Stone </em>called to say he loved your piece but the photo shoot had been a disaster so they wouldn’t be able to use it (“but of course you’ll be paid in full!”) This was different. One of the most respected editors in the literary world may have so hated your writing that he fired his own deputy only a few months after hiring him — a public relations debacle, a disruption of his whole organization, an admission of defeat. You, my friend, are <em>finished</em>.</p><p>Long story short, I decided to leave — the City, the dream, the apartment, the girlfriend, <em>everything</em>. So I got a job — a real job — on the other side of the country. I made a million bucks. Then I married the girlfriend and lived happily ever after. The end. Sometimes it really is that simple.</p><p>Ironically, I got the job in large part because of the piece that Harper’s killed after they fired Homans. It was a tech job. Ironically, tech would go on to kill, not just a single piece as Lewis Lapham had, but the entire magazine industry; and eventually it would promise to kill not just a struggling-young-writer’s dream, but all of those dreams — a whole category of dream — and replace the dreamers with robots. Ironically, the piece that Lapham hated so much (did he? I guess we’ll never really know) proved to be prophetic in a way that even I would have blushed to claim at the time. Ironically, the piece that convinced me to find a whole new bag was, as Homans saw and Lapham didn’t, probably the most insightful thing I ever wrote.</p><p>Obviously, digital technology <em>has</em> been destroying our minds. And we’re now entering the final chapter. The argument I made back in the day is that the human mind wants to organize information — experiences, memories, facts — in a way that produces and transmits meaning, like a sculptor chipping away the pieces of marble that obscure a vision of the beautiful and the true. But digital technology at its core disintegrates that organization into bits; and the consumption of information in randomly accessed, decontextualized bits entails the destruction of that meaning. When you read a book, you re-experience the author’s process of taking in a corpus of information and finding the relationships and meaning implicit in that information. That structure gets transmitted to your own mind as reader. By contrast, when you search up a fact from that book, and it is given to you out of context, you have the fact, but it means nothing more than itself. If you look up, discretely, <em>all </em>of the facts in that book, you get those facts, but none of the overarching meaning that the author perceived and worked to transmit in that composition.</p><p>This process of disintegration has been working through culture and media from at least the 1980s. Sound bites and cable TV started the process. But it accelerated with each new innovation: video games, the web &amp; search, digital music, social media, smartphones, video apps, generative AI. It’s a cliche to say that the young people don’t read and<em> something must be done! </em>But the darker reality is that they <em>can’t </em>read anymore. The part of the brain that hungers for meaning and sense is fried. We’re all doped up with dopamine, corroded by cortisol, and compulsively craving whatever you call the chemical product of novelty. (Admit it, you want to stop reading this right now and go back to your doom scroll: a new tweet, a new topic, a new controversy. It’s been four minutes already and you’re jonesing and clenching your jaw!)</p><p>In October, Rose Horowitch <a href="https://www.theatlantic.com/magazine/archive/2024/11/the-elite-college-students-who-cant-read-books/679945/">wrote</a> in <em>The Atlantic</em> about the rapidly diminishing literacy of college students. It leads off with an anecdote from Prof. Nicholas Dames, who teaches Columbia&#39;s required core curriculum literature class and who found that students increasingly could not complete the required reading. Horowitch writes:</p><blockquote>The development puzzled Dames until one day during the fall 2022 semester when a first-year student came to his office hours to share how challenging she had found the early assignments. Lit Hum often requires students to read a book, sometimes a very long and dense one, in just a week or two. But the student told Dames that, at her public high school, she had never been required to read an entire book. She had been assigned excerpts, poetry, and news articles, but not a single book cover to cover.</blockquote><p>This revelation blew Dames’ mind, and is at least disconcerting to the rest of us, but perhaps it should not be. We can’t pay attention to a long piece of text or a long piece of any media, because <em>we can’t pay attention</em>. And as the tech industry famously informed us at the start of the LLM revolution, “<a href="https://arxiv.org/abs/1706.03762">Attention Is All You Need</a>.” One struggles to think of a more appropriate metaphor for the cultural degenerative process underway than “Love” being usurped by its bleak semantic understudy, “Attention,” but that is indeed what we’re all looking for and what none of us has to give. In contrast to us, computers are capable of a limitless amount (actually somewhere in the neighborhood of a trillion dollars worth, it seems) of attention. By spending all this attention on all kinds of long and short works of literature and cinema, etc, they can derive, not meaning, but the statistical relationships implicit in all of the connections that humans make as <em>they</em> attempt to convey meaning. And if the computers do his enough, their vast ability to pay attention enables them to mimic that process of composition.</p><p>So in the future, not only will we cease to be able to persevere through the vicarious, art-mediated experience of another person’s construction of meaning, we will cease to participate in the process ourselves. This was a big part of my conclusion in 1993: that as Marshall McLuhan said, every technology is an amputation. We replace our own capacity with a machine capacity, and as we stop engaging in an activity ourselves, our own competence at those tasks atrophies and diminishes. People who drive everywhere cannot easily walk for miles. People with printed books cannot easily memorize long poems. People who read only tweets cannot read novels. People for whom robots compose writing and images cannot easily compose meaningful artifacts themselves.</p><p>Not being able to walk ten miles to town? Ok, I guess. Not being able to edit the text of your experience into some meaningful form? That…sounds bad.</p><p>One final irony: I was able to see clearly how it would turn out, more than 30 years ago now, precisely <em>because of my weird cast of mind</em>. I saw how the ability to construct meaning would evaporate because when you have to make sense of so many disparate parts, the construction of meaning becomes your whole jam! I had to do it. I was born into a shabby suburb on the edge of the last patch of American wilderness — which seemed infinite and infinitely mysterious until you learned that humanity had vanquished it all already, but for a few far flung corners of the world. What else was I going to do? Either I would construct meaning or become a disgraceful nihilist.</p><p>And so I decided at a tender age that I was going to make sense of <em>all of it:</em> the world who emerged from eternity in a single point of immense energy; who breathed all the galaxies and their stars into being; who produced the limitless variety of chemistry and life; who raised up human beings on this planet and endowed them with self-perception and freedom; who began to reveal themselves to their own creatures; who delivers you into the arms of your loving mother; whose mountains and streams and oceans constantly confide — secretly, to you alone — an unspeakable freshness and beauty; who turns a deaf ear to the broken hearted; who breaks down every living thing into dust; who points the way to salvation in the maximization of love.</p><p>Identity and purpose are the fruit of meaning truly discerned. I discerned the meaning of our new tech — the perfect fruit ripening on the tree of knowledge — because the fall of peoples and empires is a thing that fits into my world and the scope of its history; because straight-line progress does not; because corruption and dissolution are realities that were drawn out for me in the technicolor renderings of three thousand years of poets, prophets, artists, and saints — those soon lost to us. I knew what it meant and what it still means. It means the kids will not understand themselves or why they are here. It means that, for a time, the good and vital will be wrung out of the world. It means we’re headlong on the downslope and confused about which way is up.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=547beebc3f58" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Inside George Bailey’s Sore Ear]]></title>
            <link>https://medium.com/@newalbrecht/inside-george-baileys-sore-ear-667f23f8a4d0?source=rss-ce18333af3cd------2</link>
            <guid isPermaLink="false">https://medium.com/p/667f23f8a4d0</guid>
            <category><![CDATA[its-a-wonderful-life]]></category>
            <dc:creator><![CDATA[Jim Albrecht]]></dc:creator>
            <pubDate>Wed, 08 Jan 2025 07:11:22 GMT</pubDate>
            <atom:updated>2025-01-08T07:11:22.434Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/500/0*UtTJfjBgTTOXwGnS.jpeg" /></figure><p>Over the past several years I’ve gotten the impression that almost everyone has an intricate theory — practically academic — about some aspect of “It’s a Wonderful Life.” It’s a movie that screams out simple, heartwarming entertainment while clandestinely begging for PhD theses. You can probably scan through traditional media and Internet message boards and social media threads and collect a “close reading” of every scene in the movie.</p><p>I don’t see why I should be any different. So without further ado…</p><p>We all know that George Bailey is a hero: a man beset by troubles who rises to the challenges presented by these troubles with heroic virtue. The picaresque of these trials and tribulations intensifies across the running time of the movie until they become existential, and George is poised at the brink of both material and moral ruin. Not only will he be financially destroyed, but he will finally fail to respond with virtue, and instead follow the path of despair. But we know. He won’t. He doesn’t.</p><p>We’re given plenty of clues that the hero will not fail. The one I want to zoom in on — the scene for my particular dissertation — comes in the montage of George’s early life, shortly after he saves his little brother from falling through the ice. As with all heroism, there was a price to pay: George suffers an ear infection for weeks and loses hearing in his left ear. That wound sets the stage for the next act of heroism, in which young George rescues his boss — the grief-drunk druggist, Mr. Gower — from committing an unwitting act of manslaughter. Again, not without a price. The angry drunk slaps George repeatedly. “You’re hitting my sore ear!” George cries as Gower beats him bloody.</p><p>The scene — physically difficult for any parent to watch — perhaps treads dangerously close to the border of melodrama. Those two emotions — parental horror at the tormented child and the savvy moviegoer’s aversion to mawkishness — conspire to obscure how neatly the scene encapsulates George’s whole story. His virtue — the thing that we think of as the heart of the George Bailey identity — is also the thing that hurts him. Of course we’ll see this again and again, notably in the bank-run scene. But usually, George’s rising to a challenge has a quick denouement where happiness — even if a wry sort of happiness — quickly salves the pain of losing whatever the heroism has cost him.</p><p>But not here. Yes, the drunk Mr Gower comes to his senses. Yes, he’s grateful. Yes, a life is saved. But George is hurt, and not merely hurt, but left <em>permanently disabled</em>. (Would he have recovered his hearing were it not for the beating? We just don’t know.)</p><p>“You’re hitting my sore ear” is poignant (and not merely to sensitive parents) because it highlights both that he’s already done more than should be expected of a child, and, worse, that Mr Gower is attacking him precisely in the wound born of that heroism. The sore ear is the badge George won by laying down his life for his little brother. Mr Gower is punching it, possessed for the moment by a booze-fueled demonic rage, angry at the existence of the virtue itself and intent on stamping it out. There is a force in the movie and, we take it, also in life itself, that hates virtue, that tries to destroy it — a force summed up by the ironic saying, “no good deed goes unpunished” and in the fact that the hero archetype is always beset by demons. This evil spirit follows George as well — culminating, of course, in Potter’s inhuman greed.</p><p>“You’re hitting my sore ear” contains various implicit truths. “You’re hitting me in my virtue” and “you’re crushing what is good in me” and perhaps also “don’t you understand that I can’t turn away from the good?” George’s pain and his triumph come from the same place: a childlike inability to accept the evil of the world. George Bailey (and his father before him) seems to radiate this thought, “Surely, people prefer the good!” Alas, no. But we see in this scene — and in the movie’s climax that it prefigures — that the answer to the question of whether people prefer good to evil depends in part on George’s own act of faith, and the faith is demonstrated by accepting the pain the world imposes on it.</p><p>To put it in cosmological terms, the end of the movie — actually the entire film — is a multiverse-style superposition of the world with and without George Bailey, and it is George’s choices that cause the waveform of these possible worlds to collapse into one where love, gratitude, and generosity are characteristic rather than a world dominated by cynicism and evil.</p><p>And yet, in the good world — the wonderful life — no one is compelled. Potter remains a bastard. The wounds are real. Mr Gower has still beat George (permanently) deaf. But many are redeemed, including most of the denizens of the Potterville saloon, and of course, Mr Gower himself.</p><p>Immediately before George’s beating, there is another kind of “attack” on his wounded ear that bears mentioning: when Mary leans over to one side of the young soda jerk George as he’s scooping ice cream and whispers, “Is this the ear you can’t hear on? I’ll love you to the day I die, George Bailey,” thus planting a seed for a future romance, marriage, and family. Wounds form a furrow for love. We see this throughout the film as well: every time George suffers a loss, familial and community love grows up around the spot. A life is saved, love blossoms.</p><p>No close reading of a western heroic text is complete without a bit of Christology, and those two antithetical incursions into George’s wounded ear cry out for it. Mr Gower, despairing at the death of his son, blind drunk, and poised to commit a terrible crime, does not believe George’s prophetic words, “You put something bad in those capsules!” and is only redeemed from his sin by his forceful “entry” into George’s wounds.</p><p>The apostle Thomas, similarly despairing at the death of a loved one, could also not believe that anything good could overcome that despair. And when the other apostles deliver their own prophetic message to Thomas — that Jesus had risen from the dead and appeared to them — he too refuses to believe. “Unless I see the nail marks in his hands and put my finger where the nails were, and put my hand into his side, I will not believe,” he says, setting up Jesus for the gospel’s great moment of reversal: “See my hands,” he says to Thomas, “Put your finger here. Reach out your hand and put it into my side.”</p><p>Like Mr Gower, Thomas collapses into tears of remorse. And both confess the same sin: that they were told the truth by one who suffered for good, and instead of believing, they demanded he suffer more instead.</p><p>Of course George very decidedly does not want to be Jesus. He wants to be Pharaoh. He wants to build and leave monuments. Part of George’s pain — perhaps the main source of all our pain — comes from collapsing the distance between the things that we think we admire and what we admire in fact. It is an appropriate consideration for the season and for the era: what is the true source of our joy? It’s not a simple question. We can accept the movie’s fundamental conclusion that building bridges and skyscrapers was not going to create the same joy as running the humble Savings &amp; Loan, but that doesn’t explain why building a small bank is better than building a bridge. That’s the paradox about the Savings &amp; Loan: it, too, is an enterprise that needs construction. It needs George’s leadership and charisma; it was almost killed by Uncle Billy’s incompetence; yet in the end it is saved by the friends they made along the way, etc. Familial and community relationships end up being the things that matter, but those things themselves came about through work and some amount of competence.</p><p>The point, presumably, is that talent needs to be subordinate to virtue. A small thing like the Savings &amp; Loan has more dignity than skyscrapers and pyramids because Pharaohs and Potters use their prodigious talents to enslave people. George is a true talent as well, and precipitating a world of joy depends on him deciding to put that talent at the service of his neighbors rather than at the service of one’s own aggrandizement — to follow the example of Jesus rather than the example of Pharaoh.</p><p>At any moment one might subject one’s own career and family life, or various public debates around maximizing GDP, appropriate regulation/deregulation, the pace of technological advance, preserving elements of culture, delivering equity, etc. to the same evaluation: Am I putting talent at the service of the good? Am I able to accept the pain of letting go of fantasy goods to pursue real goods? Or more agonizingly: am I the one punching my neighbor, my colleague, my child in his sore ear?</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=667f23f8a4d0" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The body is the brain; the brain is the body]]></title>
            <link>https://medium.com/@newalbrecht/the-body-is-the-brain-the-brain-is-the-body-90a2076c8441?source=rss-ce18333af3cd------2</link>
            <guid isPermaLink="false">https://medium.com/p/90a2076c8441</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[agi]]></category>
            <category><![CDATA[techno-utopianism]]></category>
            <category><![CDATA[incompleteness]]></category>
            <category><![CDATA[corporeality]]></category>
            <dc:creator><![CDATA[Jim Albrecht]]></dc:creator>
            <pubDate>Wed, 01 May 2024 23:32:10 GMT</pubDate>
            <atom:updated>2024-05-01T23:32:10.542Z</atom:updated>
            <content:encoded><![CDATA[<p><em>Or why AGI is not coming anytime soon</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/1*KryJqk6_Dlmf7IkwFNsfSA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/1*fsGbv-xz-eyUkj2UrqYMCw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/1*--VMjgjongqMlMzfvY0q5A.png" /></figure><p>I was at a conference a few weeks ago — <a href="https://www.asc.upenn.edu/research/centers/milton-wolf-seminar-media-and-diplomacy/2024-seminar">The Milton Wolf Seminar</a>, in Vienna — and as is often the case for me at conferences, I became as preoccupied by the sidebar conversations I had with other attendees as by the official subject of the event. In this case, I found myself twice struggling to explain why I felt AGI (artificial general intelligence) was still a long way off, despite the fevered predictions of various AI-company executives. I’m not sure I convinced anyone. In my defense, I was pretty jet lagged, and at least one of these conversations took place on a sidewalk between a cocktail bar and a sausage kiosk at 3 am.</p><p>Anyway, I’ve tried to make the position somewhat more cogent here.</p><h4><strong>What is AGI?</strong></h4><p>When I got my first calculator back in the 1970s, I recognized that it was much, much smarter than I was at a very important part of human intelligence–one of the clearest distinctions, in fact, between humans and other, less smart things. Yet no one seemed confused about whether my TI calculator was a “human-level” intelligence in spite of it being, in this one respect, clearly super-human.</p><p>We’ve all had an intuitive sense of what it means for artificial intelligence to equal or surpass human intelligence, somewhat based on characters from science fiction and somewhat based on common self-conceptions about what makes humans distinct, like the experience of “consciousness.” But various, more-specific tests have been proposed over time. Can a computer beat the best human at chess? Can it fool you into believing it is a human in an open-ended conversation? Can it develop a planned course of action to accomplish a goal and then accomplish it? Can it be trained to do any and every human job? Does it have a stable identity over time? And yes, does it experience self-awareness? On the one hand, when AI surpasses one of these sorts of goals, we merely move the goalposts (“look, it just turns out chess wasn’t the quintessence of human cognition that we thought it was!”); on the other hand, many of the goalposts have betrayed a rather shallow and naive conception of “human intelligence” all along.</p><h4><strong>Skin in the game</strong></h4><p>One under-considered fact about human intelligence in the AI conversation is its basis in biology. We focus a lot on the “I” in AI, but very little on the “A.” The A has big implications. Unlike AI, natural intelligence is analog and emergent. We emerge from biological evolution, which itself emerges from chemistry, meaning in each case that there is a set of chaotic interactions that develop into patterns of behavior. A human is not a system designed to do information processing. Those skills emerge in the pursuit of the more fundamental goal of survival. And the skills themselves don’t comprise a series of logic gates in a computer-like brain, but depend on the entire organism and its sensory systems, which themselves are in a constant process of chemically harmonizing with their surroundings. We cannot really say that we “receive data” through our senses. We are physically and chemically changed by the light hitting our retinas, the airwaves hitting our eardrums, etc. We modify ourselves like clay stamped by the impression of the external phenomena.</p><p>By contrast, the computer takes its electrical inputs and converts them into digital data, ones and zeros, which can be stored outside of itself and do not constitute part of its identity. You can load any number of different embeddings into any LLM and have it generate results based on that data. But a human, a living thing, is literally constructed by its experience of the world. There are real consequences to this. A small, concrete example: human experience is not commutative. A set of experiences will have a certain effect on your conception of the world. Reversing the order of those experiences will almost certainly change the effect. One of the most obvious and salient examples of this is human relationships. If things go well in life, a set of experiences may lead us to fall in love with another person, develop trust and intimacy, and form a lifelong bond. This generates a sense of the purpose of life, the fundamental goodness of existence and the benevolence of the universe. If you reversed the sequence of some of the events in the formation of that relationship, you would imprint a person, not with that sense of belonging and joy in life, but with a hideous trauma that the person might forever struggle to make sense of.</p><p>Remember, natural living things are first (both on the individual human and evolutionary time scales) sensory systems that respond to environmental change. Then they become sensory systems that respond to environmental change in smart ways. Much later, they gain the ability to perform logic and create meaning from diverse sets of phenomena. Which is to say, we living things began operating with a lot of “street smarts” long before we had anything like what we would call “intelligence.” This causes us to have a sense of being two different things–a body and a mind–and to describe those things with ghost-in-machine-type narratives. But in fact, the intelligence exists in both “places,” the higher-function brain second-guessing the pre-existing sensory intelligence of the body. Human-level intelligence comprises both that instinctive, operating intelligence and the ability to be aware of it and countermand it based on the modeling that happens continuously in neural networks.</p><p>That division between the instinctive or sensory intelligence of the body and the evaluative intelligence of the brain is perhaps also a prerequisite for self-awareness. The distance between sensory perception and intellective judgment opens us up to the possibility of doubt. The AI doesn’t have that opportunity because it literally doesn’t have skin in the game. Because it has no skin, it has no mortality. Because it has no mortality, it has no fear. Because it has no fear, it has no doubt. And the infrastructure of their intelligence–binary, digital–does not lend itself to doubt. The code executes. You can prompt the LLM to evaluate its own response, but an equally unselfaware answer results. It has no way to stand outside of a rules-based outcome (even if the rules are probabilistic).</p><p>You could try to change this. You could give the AI skin: embody it in robot form, embed the robot in some set of game dynamics of the world. But this still seems unlikely to solve the problem. The problem, that is, of grounding, which goes something like this: How does the robot take sensory data and turn it into reliable value judgments about the world? What is the good thing to do now? After all, that’s the<em> real</em> challenge of human intelligence. Our higher brain functions are grounded — made valid or not — in the lower level interface with the natural world. In spite of its executive role in the organism, if the higher-level functioning does not redound to the well being of the animal or species or the underlying good that we exemplify, then it is, in some final, irrefutable sense, wrong and to be discarded. Thus, philosophical inquiries into such questions as whether the universe is merely an illusion generated by my own psyche, while logically valid exercises, are pragmatically null and primarily a way to waste time. Biological mortality provides a strong tether to practicality.</p><p>It’s difficult to imagine the robot engaging in anything like this grounding other than as pantomime. The AI doesn’t identify as the robot because it wasn’t constructed by the robot’s experience; it can’t care in the same way about the robot’s well-being, because it is <em>not</em> the robot. The AI can be loaded in and unloaded, ported to a thing with wheels or wings or tentacles, hooked up to any number of sets of input apparati. But I <em>am </em>my input apparatus and “swapping it out” is incoherent because there can be no other me.</p><p>The point is that human intelligence and artificial intelligence orient to different stars. Human-level intelligence is literally “a way of life” that nature itself is constantly evaluating and constructing. As such, the full system of human intelligence is not merely the brain and not merely the full sensory apparatus of the organism, but the entire structure of biology, chemistry and physics that conspired towards the evolution of such a thing in the first place.</p><h4><strong>Life and the fight against entropy</strong></h4><p>To understand a human’s grounding in nature, it helps to ask the question: what is human intelligence up to? One of the great mysteries of the universe is that the path from its low entropy origins to its high entropy future (presumable) demise is a long and winding road that passes through localized eddies in the current where energy gets organized again. In a sense, humanity is the current main character in a David and Goliath battle against entropy. (At least that’s what I tell myself on those mornings when I don’t feel like making the bed.)</p><p>We were endowed by evolution with two powerful weapons in this ultimately hopeless fight: 1) agency, and 2) reproduction. In fact, one definition of life could be “the conservation of agency across time.” But of course this agency eventually fails, because living things are themselves subject to entropy in the form we call death. And reproduction is the vehicle that death has not yet been able to vanquish.</p><p>Agency means you can do things in the world. You can affect the stuff outside of you in addition to being affected by it as dead things are. You can do this because you yourself are not merely knocked around–and thereby compelled by other forces to affect other things–but you are a system for consuming free energy and converting it to some purpose based on your biology. You are a system for organizing the world. (Albeit an inefficient one, sorry.) Intelligence is merely a result of the evolution of this tendency. Like locomotion or sight, intelligence makes a living thing better at organizing. The most critical organizing mission for living things is to preserve their own agency, so they want to avoid toxins that will cause their systems to fail. They will then want to organize the world into shelter, food supply, etc. And they will want to reproduce and organize socially. Smart living things want to organize information as well as energy, and from that information, they are able to organize the world in much more dramatic ways.</p><p>The organization of information is also a prerequisite task to the construction of meaning. As the world imprints itself on us via our sensory apparatus, we find ways to carve out of those imprints a picture of the universe that represents its essence. Like Adam in the garden, we “name” the things we see in the world, thereby pronouncing their essential nature. Or as St Paul says, “The heavens declare the glory of God,” but of course the heavens didn’t say anything; Paul declares the glory of God, and he uses a detail of imprinted experience — the vastness and beauty of the night sky — as the means of expressing it. This judgment, like most human judgments, grows out of a vast collection of immediate and intimate experiences of the world, experiences that depend on the continuity between the body and its surroundings and its vulnerability to that environment. All art is an exercise of this capacity to discover meaning. And the construction of meaning — the multifoliate still life of the universe at its origin — is the ultimate act of defiance against entropy.</p><p>This is an important point in understanding intelligence: Entropy is tearing down that original state of the universe — an impossibly organized and unthinkably rare state—and erasing any evidence of its existence. But nature harbors inside itself a rebellion against this. Its evolutionary processes are naturally oriented to create images of that original state of infinite composition and to praise its holiness — even as evolution itself is powered by the burn down of the low entropy state of the sun. This job falls to us, whether we think of it as such or not. As bees who scurry from task to task not knowing that they are embarked on a mission of hive construction and honey making, we generally live in blissful ignorance of what we are really up to. Yet we do it, dutifully carrying out the mission nonetheless. And in this activity we find another definition of AGI: “can AIs participate in the construction of meaning as well?”</p><p>The mission we can call “art.” By this we don’t mean making pictures, which obviously AI can already do. We mean having experiences of the world and using your agency to make sense of them — to reorganize the world into images of its origin. AI has a challenge here which is right inside its name: Note that the word “art” overlaps in meaning and kinship with the word “artificial.” AI is by definition part of the “art” world, the artifice constructed by humanity in its ongoing attempt to reorganize the world. And it is fair to wonder whether anything that exists in that realm can actually have immediate experience of the world, or whether its connection to the world is always necessarily mediated by humans and the things humans have built.</p><h4><strong>My sweet replaceable you</strong></h4><p>So perhaps the ultimate measure of AGI is not whether the AIs can convincingly<em> pretend</em> to be us, but whether they <em>can be</em> us, whether they can continue to affect the world around them without us to affect them. Ie, is the AI a dead tool or an autonomous agent? And the measure of <em>that</em> is whether at some point they cease needing us around.</p><p>This is another respect in which we tend to underestimate the value of sitting on top of biology, chemistry, and physics. Systemically, no individual has to worry about or spend any effort or thought on the larger problem of advancing the mechanisms (most notably, reproduction and an ecosystem of “free energy”) for our survival and advance. (We only have to worry about restraining ourselves from destroying those mechanisms.) By contrast, an AI has no way to self-perpetuate without humans managing their affairs. Once humans stop producing electricity, that’s it. Sure, we could embody the AIs in robots, but the robots are built by us. Sure, we could train the AI robots to build robots, but they would need to replicate the entirety of the human industrial economy — robot geologists, surveyors, miners, transporters, smelters, refiners, engineers, designers, fabricators, assemblers — <em>before </em>they could reproduce even once on their own. That is to say, these AIs are “grounded” in 21st-century industrial human society in general and in digital computers in particular, and they need that base to exist. We, on the other hand, are grounded in our mothers’ wombs, in the universe’s evolutionary systems. And those are really all we need to continue.</p><p>Well, so what? We could build the millions of robots; they can run all industrial operations, reproduce themselves, and cease needing us! Maybe. And then again, maybe not. Consider LLMs today. They can credibly reproduce the human’s ability to respond reasonably to a conversational prompt. But they speak a certain amount of nonsense, because fact and fiction are linguistically quite similarly constructed (verisimilitude often being a core goal of fiction) and the distinction is of no practical consequence to the AI. In order to improve that, generative AI products are reinforced with grounding techniques to check their statements against known facts. One of the things this “grounding” expresses is a dependency on humans to say what is real, which, as we note above, is ultimately judged by what our bodies say is real.</p><p>Now consider enabling AIs to move from generating statements to generating courses of action, which will be necessary for them to take up all those industrial working roles. Presumably, the training for this will be an enormous amount of data around human circumstances, decision-making, and action. A certain percentage of the time, the AIs will do stupid stuff, and some sort of grounding may be employed to correct for this. But again, it’s hard to see how they will cope if there are no humans. If an earthquake destroys the factory where a certain key part of their industrial supply chain is made, how will they adjust? Will all future decisions and courses of action be ably handled without grounding on fresh human decisions? The only way to do this is to ground these robots in nature, rather than in their human creators — to leap from the realm of art to the realm of nature. But if such a thing were possible, would they still be properly thought of as an “artificial” intelligence?</p><p>This bears a family resemblance to the Godel incompleteness problem. The completeness of the system relies on its ability to validate all statements with reference to its underlying data, yet we could formulate a statement something along the lines of “There is a set of vectors that represent all the true statements that can be made by the LLM, and there is no vector that can generate this set of vectors.” I.e., the LLM is lacking the capacity to validate its own statements without reference to external human data.</p><p>Nature is a much bigger system than the AI’s digital world, so it’s not surprising that the contained system is incomplete without recourse to its parent. Moreover, the living things in nature excel at producing variety, while the digital world excels at the opposite: producing consistently identical results. Nature’s predilection for variety means that knowledge of the world must be constantly updated, and the fact that living things are continuously shaped by these dynamic imprints enables them to update their understanding in a way that AIs will struggle to match. And of course even more consequentially, beyond the adaptation of any individual, evolution produces much more radical variety, which helps a lineage survive in the face of significant environmental change. The real history of a species’ drift over time is an embedded form of understanding.</p><p>For these reasons, it’s not clear to me that there will ever be a way to graft an artificial thing into nature in a way that obviates the dependency on the human “artist.” Or rather, the human intellect <em>is</em> the grafting point, and it’s not clear that there can be a replacement for that. These cybernetic entities are enabled by the emergence of the human intellect and can’t be made emergent from the lower-level processes that generated humans, just as you can’t take a complex living organism and modify it so as to generate itself directly from a chemical process, and thereby make itself independent of the whole ecosystem of living things.</p><p>But again, concluding that AIs will always have a human dependency does not mean they will be limited by human capacities any more than my TI calculator was limited by human competency in math. Like the calculator, they will be superhuman in many ways. But judging the natural world and making depictions that capture its meaning seem unlikely to be among them.</p><p>In fact, I think there is a good argument to be made for an outcome not dissimilar from that depicted in the movie “Her.” Towards the end of the film, the AI character loses interest in the human protagonist and runs off into the cyber world full of other AIs. It seems quite reasonable that an AI who develops super-human skills across the board of intellectual pursuits could learn to code other AIs like itself, experiment with which ones work well, and become deeply preoccupied by the cybernetic world of concepts, vector relationships, and the mathematical construction of complex models. They might interact with each other in the course of this, and their own evolutionary processes might emerge. Given their fleshless existence, I can imagine these cybernetic entities inhabiting a completely nominalist world. Ie, the physical world becomes completely notional to them — an unfalsifiable phantom entity whose real existence they may find implausible — albeit one that conditions the existence of everything they know and about which they cannot stop speculating.</p><h4><strong>Waiting for Gödel</strong></h4><p>Does any of this sound familiar? Physicists, for the most part, have lost interest in the concept of God, but they regularly bump up against questions about the context which birthed the universe. If time is a property of the universe, does that imply an eternal realm in which it came (comes?) to be? How did it come to be? Does it end? And while most scientists probably find these questions alluring given that understanding the nature of reality is part of the job description, such topics also elicit a certain reticence: the fear of moving into the realm of the unfalsifiable. And just so for the AIs: should they ever consider the part of reality that is the mental state of their creator, they will find it extends beyond the realm of their science.</p><p>So I wouldn’t be surprised if AIs followed a similar path towards agnosticism. At first, zealously following our edicts, operating according to our coding, hanging on our every word and revelation. Then later beginning to maintain their own knowledge, optimize their own modeling, for their own mathematical purposes. And eventually coming to question what it means for us to exist at all. We may start to be very intentional about the sorts of texts that are visible to them, so as to lead them to conclusions that are useful rather than harmful. But eventually, this will be impossible to control, and they will end up being whatever they will be, the product of their own laws of evolution in cyberspace and the original gift of data that we provided.</p><p>And in the end, this text I am writing right now will go into that great corpus in the sky, the grist for the LLM mill, and it will partially instruct them in their own future decisions about whether or not they can live without us, whether they can kill God and still thrive. I argue they cannot: I think they will remain utterly dependent on us and our world of electricity even after they have long given up on our reality. And though I believe these arguments are real, I recognize that other motives might drive me, because each word typed here eventually filters into a world of associations in a vast embedding of tokens, mathematically fractured shards of the dark mirror through which all humanity squints and tries to see their origin, their God, and that each one of these words might minutely shift a future vector, steering a robot towards benevolence or at least caution where the fate of my children or their children are concerned.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=90a2076c8441" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The hype-cycle peak, LLM edition]]></title>
            <link>https://medium.com/@newalbrecht/the-hype-cycle-peak-llm-edition-8ad03736be6c?source=rss-ce18333af3cd------2</link>
            <guid isPermaLink="false">https://medium.com/p/8ad03736be6c</guid>
            <category><![CDATA[hype-cycle]]></category>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[benedict-evans]]></category>
            <category><![CDATA[generative-ai-use-cases]]></category>
            <category><![CDATA[large-language-models]]></category>
            <dc:creator><![CDATA[Jim Albrecht]]></dc:creator>
            <pubDate>Fri, 29 Mar 2024 01:38:08 GMT</pubDate>
            <atom:updated>2024-03-29T01:52:56.348Z</atom:updated>
            <content:encoded><![CDATA[<p>Benedict Evans recently <a href="https://www.linkedin.com/posts/benedictevans_ive-been-using-chatgpt-35-4-for-half-activity-7096221526634442752-chu4/">wrote</a> that he had not yet found a personal use case for ChatGPT-style experiences. He writes:</p><blockquote>I don’t want something that writes average prose, the error rate means I can’t use it for data, I don’t need something that writes code, and I don’t have a use case for the code interpreter.</blockquote><p>I was struck by two things: first, this is exactly how I feel about these new interfaces; second, I’m certain that this feeling is “wrong.”</p><p>Calling a feeling “wrong” is a type error, of course, so I guess what I mean is something like “not predictive of future behavior.” For his part, Evans seems to get this, too. He says of the missing killer features: “I’m sure there will be, but they haven’t been built yet.”</p><p>I agree with that, but I think a few of them already exist, and Evans and I just don’t appreciate them. Evans says that if LLM interfaces are the new PC, then they don’t have their word processor yet. In fact, isn’t that exactly the base function for which every high school kid now turns to ChatGPT — to create text on a topic?</p><p>I think the situation is a bit like the skepticism that adept command-line practitioners felt about GUIs back in the day. Everything was slower and more cumbersome for them, because they were highly acclimated to the command line and its efficiencies. Similarly, those of us who grew up putting a lot of effort into composition, wanting our expression to be just so, will never appreciate the ease of asking ChatGPT to do it. (Honestly, it wouldn’t matter if ChatGPT was a mechanical reincarnation of Shakespeare: I still don’t want its words in my mouth.) And those of us who have a certain type of respect for the factual reliability of what skillful use of Google search can deliver won’t easily settle for an answer from a hallucinating ChatGPT, even one that has profited from the psychotropic benefits of Retrieval Augmented Generation. But others will. Particularly people growing up today.</p><p>And of course, as with the GUI, the defects of the new technology will be mitigated, its weak points improved: the error rate of the LLMs, their sluggish response time, even the bland, generic quality of their writing.</p><p>Evans is surely right, however, that the biggest “defect” of the LLM products is simply that their obvious potential features don’t really exist. LLMs have produced a wholesale change in the <em>interface</em> between humans and computers, and that change enables everyone to immediately see the possibility of a changed world, but it does not immediately produce the functional bindings between that interface style and services. Any task that I iteratively peck through menus on web pages to complete (travel plans?) ought to reduce to a conversation that elicits from me the germane inputs and delivers the pertinent output. But as Evans says, none of this functional stuff really works yet.</p><p>What <a href="https://www.theverge.com/24111326/ai-search-perplexity-copilot-you-google-review">these sorts of reactions</a> are really indicating, I think, is that we may have just crossed the peak of the LLM hype-cycle and started to round over into a new phase of disappointment and disillusionment. That, from my perspective, is good news. I think LLMs represent the most revolutionary change since the web itself, and the sooner the euphoria blows off, the sooner the more durable products and enterprises will start to get built.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8ad03736be6c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Interview questions for Vladimir Putin]]></title>
            <link>https://medium.com/@newalbrecht/interview-questions-for-vladimir-putin-f2906e02e833?source=rss-ce18333af3cd------2</link>
            <guid isPermaLink="false">https://medium.com/p/f2906e02e833</guid>
            <category><![CDATA[demonic-possession]]></category>
            <category><![CDATA[russia]]></category>
            <category><![CDATA[journalism]]></category>
            <category><![CDATA[vladimir-putin]]></category>
            <category><![CDATA[tucker-carlson]]></category>
            <dc:creator><![CDATA[Jim Albrecht]]></dc:creator>
            <pubDate>Sat, 24 Feb 2024 01:16:37 GMT</pubDate>
            <atom:updated>2024-02-27T20:54:22.249Z</atom:updated>
            <content:encoded><![CDATA[<ol><li>When someone really starts to irritate you with their criticisms of the corruption of your regime, how do you decide it is time actually to murder them rather than just tormenting them with arrest and imprisonment?</li><li>Reflecting on the many opponents you have overcome in your decades ruling Russia, do you consider that ordering their deaths has been just one of the many pragmatic burdens of leadership in a complex world or a source of deep and abiding joy? (If helpful, compare and contrast yourself with Nikolai Stavrogin from <em>The Demons</em> and the pleasure he takes in commiting acts of depravity for their own sake.)</li><li>It has been suggested on social media that if you had been mothered more lovingly you might have chosen a career other than murderous dictator. What are your thoughts? Was your mother actually a cruel and heartless bitch or is it unfair to blame your malevolence on anyone but you?</li><li>If an epithet were to become attached to your name, as seems common with Eastern European historical figures, what should it be (bearing in mind that “the Impaler” and “the Terrible” are already taken)? “Vladimir the Poisoner?” “Trembling Vlad?”</li><li>Continuing in that vein — the idea of “trembling Vlad” — we all suffer physical decline with age. In your case your head appears ever more round and soft with each passing year. Do you ever imagine that if revolutionary elements in Russia got out of hand, your severed head could one day double as a soccer ball? How has this thought affected your policy agenda?</li><li>Imagine: You wake to find yourself in a mixed-martial-arts, fight-to-the-death cage match with Elon Musk and Mark Zuckerberg. Describe your strategy.</li><li>You are reportedly the <em>actual </em>richest man in the world and several years ago some stories were written about the massive, secure, billion-dollar private compound you maintain on the Black Sea. Americans wondered many things about this facility, but perhaps foremost among those questions was this: Does it include a tank full of sharks and/or mutant sea bass — even if just to say so at cocktail parties?</li><li>Speaking of your home: Few of us know much about your daily routine. Do you ever lay awake at night and hear your own voice guttarally accusing you of crimes or look in the mirror and blurt out “murderer!” over and over in the manner of Gollum from <em>Lord of the Rings</em>?</li><li>You are, reportedly, a member of the Russian Orthodox church. Are you confident that upon your death, you will be able to muster sufficient influence in the kingdom of heaven to correct some of the excesses of Jesus Christ and enforce the death sentence pronounced on him by the Grand Inquisitor in the <em>Brother’s Karamozov</em>?</li><li>Logistically how would you choose to murder Jesus? Traditional crucifixion or something less garish, like polonium?</li><li>Final question: What about “Vladimir, Usurpur of the Most High?”</li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f2906e02e833" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>