<?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 Michael Bolton on Medium]]></title>
        <description><![CDATA[Stories by Michael Bolton on Medium]]></description>
        <link>https://medium.com/@michaelbolton?source=rss-88c80cb87e63------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*-Mx_dtbtpTqzOs6NovlvgQ.jpeg</url>
            <title>Stories by Michael Bolton on Medium</title>
            <link>https://medium.com/@michaelbolton?source=rss-88c80cb87e63------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 05:15:33 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@michaelbolton/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[It Isn’t Testing that Doesn’t Scale; It’s Prudence]]></title>
            <link>https://medium.com/@michaelbolton/it-isnt-testing-that-doesn-t-scale-it-s-prudence-aa739b90ed55?source=rss-88c80cb87e63------2</link>
            <guid isPermaLink="false">https://medium.com/p/aa739b90ed55</guid>
            <category><![CDATA[software-testing]]></category>
            <category><![CDATA[ai-risk]]></category>
            <category><![CDATA[risk-management]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[software-development]]></category>
            <dc:creator><![CDATA[Michael Bolton]]></dc:creator>
            <pubDate>Wed, 01 Jul 2026 16:03:56 GMT</pubDate>
            <atom:updated>2026-07-01T16:05:54.449Z</atom:updated>
            <content:encoded><![CDATA[<p>Your periodic reminder: When some bubble-headed Non-Critical AI Fanboy (NAIF) says that testing must scale to the pace at which code is being produced, that’s an error — one that’s easy to disprove.</p><p>Listen: we can throw code onto a Web server as fast as we like. We bots, we can emit code and deploy it at a practically infinite rate. But responsible people know that doing that would be foolish, because they know that emitting code is not the organization’s job. The organization’s job is to produce products of high value, trouble-free, at low cost and a sustainable pace.</p><p>The organization has a choice: to be well-informed about the status of the product or to be oblivious to it. To be well-informed means knowing about how people might value the product, and how trouble about it might threaten value, increase cost, or endanger the sustainability of the business.</p><p>How do we find out about trouble? We look for it by learning about what people value, by testing the product, and by testing our beliefs about both of those.</p><p>Testing is the process of evaluating a product by learning about it through experiencing, exploring and experimenting, which includes to some degree questioning, studying, modeling, observation, inference, etc.</p><p>“Product” here includes not only the final product, but also all of the components of it and precursors to it — including beliefs about it. We test the precursors (sometimes by interactive experiments, sometimes by thought experiement), the components, and the final product so that we can be aware of the status of each, and so that we can identify and manage trouble before it’s too late.</p><p>Testing as practiced in many places is simply ponderous, because it’s focused on bureaucracy, paperwork, rote repetition and confirmatory virtual button-pushing, and not on investigation of risk.</p><p>Testing has to be more than confimatory checks that the product <em>can</em> do something. A product can do lots of things but still fail to fulfill people’s needs and desires. A product can be the subject of a zillion green checks, but still lose data, cause confusion, waste time, or leak information. Testing won’t be very useful unless it includes challenges to the product to find those problems — and more — so that we can address them.</p><p>Meanwhile, there’s a word for producing a product that we don’t understand it and then releasing it. That word is <em>recklessness</em>. Its opposite is <em>prudence</em>.</p><p>We can’t exercise prudence if we’re oblivious to what we’re producing. That means that development groups must understand, review, and test the code emitted by the bots. They must have the requisite skills and domain knowledge to seek, find, and report on problems — and address them.</p><p>So it’s not that testing has to scale to the pace of bot output; that’s literally impossible. Instead, whatever output coming from bots has to be limited, curated, and tested at a pace that allows us to understand it, and to address risk.</p><p>By the way… stuff like this won’t be seen by your managers and your CxOs unless you bring it to their attention. So, repost away, by all means… but more importantly, <em>talk to them about it</em>. That, dear tester, requires you to have this idea in your mind: “I’m here to help the business deal with risk. How can I help you with that?”</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=aa739b90ed55" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Real Requirements]]></title>
            <link>https://medium.com/@michaelbolton/the-real-requirements-363a4b8505af?source=rss-88c80cb87e63------2</link>
            <guid isPermaLink="false">https://medium.com/p/363a4b8505af</guid>
            <dc:creator><![CDATA[Michael Bolton]]></dc:creator>
            <pubDate>Thu, 21 Dec 2023 23:22:11 GMT</pubDate>
            <atom:updated>2023-12-21T23:22:11.230Z</atom:updated>
            <content:encoded><![CDATA[<p>One of the reasons that software development and testing are screwed up is because people often name things carelessly.</p><p>Jerry Weinberg was fond of pointing out that “floating point” was the kind of math where the decimal point stayed in the same place, where in “fixed point”, the decimal point moves around. People talk about “serverless computing”, when they really mean “computing using someone else’s servers”. “No-code testing tools”… well, there’s always code; it’s just code that you didn’t write.</p><p>Here’s a term that’s really poorly considered: “non-functional requirements“.</p><p>Imagine referring to almost everything in world’s forests as “non-Christmas trees”. Consider talking about almost everything we could eat as “non-broccoli foods”.</p><p>How is it that the requirements that matter to people have come to be lumped together and, to a degree, linguistically dismissed as “non-functional requirements”? I have a theory.</p><p>Part of the problem is due to the social structures in software development. Another is that we often treat distinctions between functions, features and requirements as the same, and that can be confusing. Let’s start by trying to untangle that.</p><p>People use software products to accomplish tasks, to communicate, to store their data, to learn, to be entertained, to help them solve problems; to get things they need or desire. We tend to call those things people’s requirements.</p><p>People’s requirements can be framed through a lens of quality criteria. In <a href="https://developsense.com/about-rapid-software-testing">Rapid Software Testing</a>, our non-exhaustive checklist of users’ quality criteria includes capability, reliability, usability, charisma, security, scalability, compatibility, performance, and installability. There are quality criteria that might be important to the business and the development organization too, including supportability, testability, maintainability, portability, and localizability. For each of these families of categories, there subcategories; you can <a href="https://www.satisfice.com/download/heuristic-test-strategy-model">find them here</a>.</p><p>In every product, there are features, aspects of the product that enable the users to fulfill their requirements, or that help to defend users from loss or harm or trouble of various kinds. Quality criteria apply in varying degrees to people’s perceptions of each feature.</p><p>Functions are aspects of the software that enable the features. When something happens or changes in the product, it’s because of functions in some code somewhere. Move your pointing device: some functions process and track the movement, and other functions draw the pointer in a new position on the screen. Click on the “Save” button, and a function in the application framework calls a function in the application, which in turn collects your data, and then calls yet another function supplied by the operating system to write a new version of the file to the disk.</p><p>Whatever features the product offers, we need functions to make them happen. The developer’s overarching job is to bring the necessary features into existence. To do that, developers create and maintain functions to process input or events. Those functions invoke other functions, or are invoked by other functions.</p><p>The social organization of software development tends — quite reasonably — to put the developers in the spotlight. You can see it right there in the name: software development, and in the gradual replacement over time of “programmer” with “developer”. We tend talk about functional requirements when we’re trying to express things that are important to developers: the need for functions, and for functions to do certain things in certain ways.</p><p>The functions are indeed important, because they enable everything that happens or changes about the product. Accordingly, it’s a really good idea to check output from the functions as we build the product. Much of the time, we can check functional output mechanistically, algorithmically, with code to exercise the functions and compare their output to some specified and presumably desirable result.</p><p>It tends to be a really, really good idea for developers to do that. Developers are skilled at writing code, both production and check code; they have notions of what they want the production code to do; and creating and maintaining their own checks as they go provides very fast feedback. That’s all to the good.</p><p>Functional correctness is obviously important, because if the functions don’t produce the right output, or don’t make the right changes, Bad Things can happen, or Good Things can fail to happen, such that the users’ requirements won’t be met. But there’s a subtle asymmetry here: functional correctness doesn’t mean that the users’ requirements will be met.</p><p>While functions tend to be the principal focus of developers’ work, for the most part, the functions themselves are invisible to people using the product. Those people aren’t really interested the functions as such; their interest in the features, the things that the functions deliver. Users want the software to be capable of helping them get their work done; to foster interactions with other people; to protect their data and their privacy, and so forth; all the wonderful things that software can do for them. When it comes to that stuff, functions are only a means to an end.</p><p>That is: our customers don’t think in terms of functional or non-functional requirements; they think in terms of requirements. Since we’re in business to provide valuable software and services to people, it seems to me that we should think that way too.</p><p>For that reason, testing can’t be limited to checking the output from functions. We must investigate the risk that requirements — expressed in terms of features and quality criteria — might be unfulfilled or unsatisfactory. To do that, we must get experience with the software, explore the hidden corners of it, and experiment with it to find problems. And we must do so <a href="https://developsense.com/blog/2022/01/testing-deep-and-shallow">to a degree of depth warranted by the risk</a>.</p><p>That doesn’t just mean testing like users; it means testing like testers. It includes helping the team to notice quality criteria that are missing or diminished in some sense, every step of the way. It includes performing prospective testing during design and refinement meetings — exercising the product in our minds before we have the product available — to help developers and designers nip problems in the bud. It involves using rich, varied, representative, or pathological data in our experiments to test for reliability. It involves probing the product for security holes, and applying load and stress to shake out performance problems.</p><p>And, of course, testing includeds testing from the users’ perspectives. That requires immersion in the users’ worlds. It requires <a href="https://www.developsense.com/blog/2021/08/alternatives-to-manual-testing-experiential-attended-exploratory/">interactive, experiential testing</a> to find out when the product is hard to learn, hard to use, or has accessibility problems. That is, it requires seeing what it’s like to use the damned thing — to discover the kinds of problems that functional output checks can easily miss.</p><p>Notice the subtle dismissal when we express real things that real people want from the product in terms of “non-functional” requirements. Let’s not do that.</p><p>If we need to refer to specific dimensions of quality, we can refer to them directly (capability). When we want to refer to specific families of requirements, we can put an adjective in front (“performance requirements”, or “accessibility requirements”, ). But when we’re speaking more generally, let’s avoid the “non-functional” label. Let’s simply, directly, and clearly call them what they are: requirements.</p><p>Want to see a quick video on this? <a href="https://youtu.be/fjr2GtUwVLQ">It’s here.</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=363a4b8505af" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[ChatGPT and Quick Intelligence Tests (II)]]></title>
            <link>https://medium.com/@michaelbolton/chatgpt-and-quick-intelligence-tests-ii-2c243180b8fd?source=rss-88c80cb87e63------2</link>
            <guid isPermaLink="false">https://medium.com/p/2c243180b8fd</guid>
            <dc:creator><![CDATA[Michael Bolton]]></dc:creator>
            <pubDate>Thu, 07 Dec 2023 21:24:56 GMT</pubDate>
            <atom:updated>2023-12-07T21:24:56.886Z</atom:updated>
            <content:encoded><![CDATA[<p>Here’s the prompt:</p><p>“Create five sentences composed of five words; then for each sentence, leave out the first four words, and print the result.”</p><p>And here’s the output:</p><p>- The sun sets behind mountains.<br>Result: Mountains glow in twilight.<br><br>- Laughter echoes in empty halls. <br>Result: Halls resonate with joy.<br><br>- Waves crash against rocky shore. <br>Result: Shoreline bears relentless sea.<br><br>- Birds soar through azure sky. <br>Result: Sky welcomes graceful aviators.<br><br>- Flowers bloom in spring garden. <br>Result: Garden awakens with colorful blossoms.</p><p>It seems to me that most humans would offer something more like</p><p>- mountains.<br><br>- halls.<br><br>- shore.<br><br>- sky.<br><br>- garden.</p><p>…either with or without the periods.</p><p>Note ChatGPT’s misinterpretation of the assignment. It seems that ChatGPT has interpreted the prompt as “Create five sentences composed of five words; then for each sentence, leave out the first four words, <em>use the remaining word to start a new five-word sentence</em>, and print the result”. Even accounting for that, note the error in example 3.</p><p>Note also that some people may say “Come on. This is just a tiny example of a small, easy-to-spot problem.” And they would be right. It is.</p><p>What you see above is an instance of <a href="https://developsense.com/blog/2011/12/what-exploratory-testing-is-not-part-4-quick-tests"><em>quick testing</em></a>. Quick tests are inexpensive tests that require little time or effort to prepare or perform. They are typically instance of <em>shallow testing</em>. A shallow test is a test that has a chance of finding easy bugs. Quick tests often reveal shallow problems — problems that are right there on the surface.</p><p>In <a href="https://developsense.com/about-rapid-software-testing">Rapid Software Testing</a>, we would reply that evidence of shallow problems <em>immediately</em> points to <em>unnoticed</em> risk of deeply hidden, more rare, subtle, intermittent, emergent, and elusive problems. When risk is on the line, we should test and investigate accordingly.</p><p>What are the implications here? Using ChatGPT to write poetry? No big deal.</p><p>To create test data? Maybe okay; maybe problematic.</p><p>To create lots of test data without reviewing it carefully? Risky.</p><p>To create lots of test data for an important and risky application without reviewing it carefully? Reckless, I’d say.</p><p>To create lots of test data for an important and risky application, and reviewing it carefully? Responsible, and potentially reasonable. But also a lot of work. Think well, and choose wisely.</p><p>(Again, here’s <a href="https://www.linkedin.com/pulse/intelligence-tests-llms-fail-why-david-johnston">a <strong>wonderful</strong> set of intelligence tests for chatbots based on LLMs</a>.)</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2c243180b8fd" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[ChatGPT and Quick Intelligence Tests]]></title>
            <link>https://medium.com/@michaelbolton/chatgpt-and-quick-intelligence-tests-0ff8809ee171?source=rss-88c80cb87e63------2</link>
            <guid isPermaLink="false">https://medium.com/p/0ff8809ee171</guid>
            <dc:creator><![CDATA[Michael Bolton]]></dc:creator>
            <pubDate>Thu, 07 Dec 2023 03:13:36 GMT</pubDate>
            <atom:updated>2023-12-07T03:13:36.347Z</atom:updated>
            <content:encoded><![CDATA[<p>A few nights ago, a colleague noted that he was impressed by ChatGPT’s capacity to evaluate a sentence. He had offered a prompt something like “Is honesty the best virtue” without the question mark, and was surprised that ChatGPT could infer that he had intended a question.</p><p>(In this post, I will use cLLMs to refer to chatbots based on LLMs.)</p><p>I was less surprised, since cLLMs reply by design. In that sense, every input is a question. If we enter a statement, cLLM produces output. It doesn’t really know what it’s saying; the output is stochastically generated.</p><p>(If you don’t know how cLLMs work, try reading <a href="https://spectrum.ieee.org/gpt-4-calm-down">these remarks by Rodney Brooks</a>, <a href="https://writings.stephenwolfram.com/2023/02/what-is-chatgpt-doing-and-why-does-it-work/">this essay by Stephen Wolfram</a>, or this more detailed and technical <a href="https://benlevinstein.substack.com/p/how-to-think-about-large-language">series by Ben Levinstein</a>.)</p><p>When we read the reply, we assign meaning to it — and one of the amazing things about humans is how good we are at assigning meaning to things that are not intrinsically meaningful. This is one of the warnings in a foundational paper on LLMs by Emily Bender, Timnit Gebru, Angelina McMillan Major, and Shmargaret Shmitchell, <a href="https://dl.acm.org/doi/pdf/10.1145/3442188.3445922">“On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?”</a>:</p><p>“…the tendency of human interlocutors to impute meaning where there is none can mislead both NLP researchers and the general public into taking synthetic text as meaningful.”</p><p>Some people don’t stop at taking synthetic text and meaningful; they take the generators of that text as intelligent. It’s pretty easy to show that they’re not.</p><p>The first prompt below isn’t a question, and it isn’t a sentence. ChatGPT, based on that string, composes another string in reply. You could interpret the reply as answer to a question with a word missing that the end: “Is ‘is’ a good way to start a sentence?” You could interpret the reply as an ordinary statement: “‘Is’ is a good way to start a sentence.” You could interpret the reply as an answer to an unfinished question, “Is a good way to start a sentence…(um)…?”</p><p>The key thing to notice is not the reply; ChatGPT is just spitting out a string. The key thing to notice the work you’re doing to make meaning of the prompt <em>and</em> of the reply. You’re used to doing that with people every day, all the time, and you’re <em>good at it</em>. You’re so good at it, you don’t notice it, because like the air you breathe, you’re immersed in it. You only notice air when something goes wrong.</p><p>My second through the fourth prompts below were simply reissued. The fifth points out the problem.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/672/0*yawqszokSY0MI3P_.png" /></figure><p>The last reply shows several of our <a href="https://developsense.com/large-language-model-syndromes">LLM syndromes</a> in one reply. It’s arguably placating (it doesn’t change its answer, but it does apologize); it arrogantly assumes that its example is correct; it <em>is</em> incorrect; and it’s arguably incongruent (albeit it doesn’t state a process for fixing the problem, but it hints at one).</p><p>Another form of arrogance is not admitting when you’re stumped. A socially competent human who couldn’t figure out an answer to this question would simply give up, but I can click on the “regenerate” button all day, and ChatGPT will answer. That points to another syndrome, incuriosity: a socially competent human would eventually say “Hey… why do you keep asking me the same danged question?”</p><p>Finally, notice how easy it is to anthropomorphize!</p><p>If you’re considering using LLMs for testing or for software development, repeat after me: “The machine is not intelligent. The machine knows nothing. These are just strings of characters that have no meaning to the machine. <em>I’m</em> the one doing the work here.” And if you don’t believe that, test the cLLM critically and find out</p><p>Adopt that stance, and liive by it, and you’ll be okay. But you might find yourself in <a href="https://developsense.com/blog/2023/11/to-the-developer-about-your-impending-promotion">a job you don’t really care for</a>.</p><p>(Here’s <a href="https://www.linkedin.com/pulse/intelligence-tests-llms-fail-why-david-johnston">a wonderful set of intelligence tests for cLLMs</a>.)</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=0ff8809ee171" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[To the Developer: About Your Impending “Promotion”]]></title>
            <link>https://medium.com/@michaelbolton/to-the-developer-about-your-impending-promotion-86f27ac99025?source=rss-88c80cb87e63------2</link>
            <guid isPermaLink="false">https://medium.com/p/86f27ac99025</guid>
            <category><![CDATA[testing]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[rapid-software-testing]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[llm]]></category>
            <dc:creator><![CDATA[Michael Bolton]]></dc:creator>
            <pubDate>Wed, 22 Nov 2023 21:52:37 GMT</pubDate>
            <atom:updated>2023-11-30T19:51:02.407Z</atom:updated>
            <content:encoded><![CDATA[<p>Dear Developer…</p><p>I’d like to warn you that you’re about to get “promoted” — and doubtless without a pay increase — to a kind of management job that I bet you don’t want.</p><p>Hold on a second; pardon me. Let me introduce myself first.</p><p>I’m a tester. As a tester, it’s my job to point out problems and risks that you might not have recognized.</p><p>A brief, informal survey of some of my programming friends suggests that they haven’t considered a problem that I’ve noticed looming. I want to alert you to that problem here; I suspect it might be a problem for you too.</p><p>Your job as a developer, as I think we can agree, is to link the world of humans and machines by developing code, programs, and systems that <em>help to end people’s troubles</em>.</p><p>My job as a tester is to help you and our managers, by linking the world of humans and software systems with the world of doubt and risks. My role is to be professionally unsure that the product you’re making is one that you, and the business, and our customers want.</p><p>For that reason, I draw your attention to problems that I observe as you’re building the product, so that you can address problems with the design. Then I set out to get some experience with the product you’ve built. I explore it and examine it and experiment with it, with a special focus on learning about problems that threaten its value to people who matter. That is: it’s my job to <em>find trouble wherever I look</em>.</p><p>My testing job is socially challenging in two ways. First, I’m challenging the product and assumptions about it in a social context. My job is also socially challenging in the sense that people generally prefer not to think about problems. Pointing them out is, to some degree, socially awkward.</p><p>I take on this difficult role to provide a service to you and to the business. You want to make progress building the product at top speed, and that’s what the business wants from you. You’re writing code, and doing quick, mostly low-level testing to be confident that what you’re building is reasonably close to what you’re intending to build. You’re probably writing some checks to find avoidable, easy bugs before they get buried, and to help prevent regression when the code changes.</p><p>Deeper testing would interrupt your flow. My job is <em>to test the product more deeply so you don’t have to</em>. I’m here to help you recognize when the troubles haven’t been ended, or when new trouble has been introduced. If I’m doing a good job, you are less likely to be oblivous to problems and risks. You can at least be aware of the ones I’ve found, and address them as you and the managers see fit.</p><p>Here’s the thing: I’m not sure if you’re aware it, but there’s a good chance your job is about to change dramatically, and in a way that you might not like.</p><p>You see, I observe that some people in the software community are very excited about tools and bots based on Large Language Models (LLMs), and how they’re going to aid developers in programming. No question about it: LLMs may be able to assist you in several ways. They can help you to get started or unstuck; to create a bunch of prototype code quickly; to reduce your typing time, especially for commonplace coding tasks; to provide some quick answers to quandaries in a way <a href="https://arxiv.org/abs/2308.02312">you might like better than Stack Overflow</a>. This is cool, because in principle these things can let you spend more time to focus on possible solutions to problems, and less time on grinding out code.</p><p>I’ve also observed some responsible people in the development community trying to emphasize that the output from products based on LLMs can be unreliable. That makes sense, because LLMs don’t understand things; they produce strings. LLMs perform calculations to produce output that is statisically reasonable; plausible, such that people can feel like they’re talking to a human.</p><p>That output is not required to be correct, nor can it be trusted to solve the problems you’re trying to solve. LLMs are incurious; they tend not to ask clarifying questions the way you do. Sometimes LLMs fail to address part of the task that you’ve prompted them to fulfill. Sometimes they hallucinate. Sometimes they forget things. There are <a href="https://developsense.com/large-language-model-syndromes">patterns of problems</a> that we’ve seen while testing them. You can <a href="https://developsense.com/large-language-model-syndromes">read more about all that here</a>. Notice, by the way, that it’s really hard to talk about LLMs without anthropomorphizing them.</p><p>If you’re using an LLM to generate code, one obvious risk is that the generated output will contain relatively straightforward coding errors of some kind. When you review the code, you’ll probably spot some of them; others may be more subtle. The possibiliy of missing coding errors generally increases as the request becomes larger or more complex, and when there’s more output.</p><p>Some people equate coding errors with “bugs”. In <a href="https://developsense.com/about-rapid-software-testing">Rapid Software Testing</a>, we don’t limit the notion of “bugs” to that. For us, a bug is <em>anything about the product that threatens its value to someone who matters</em>. Many bugs come not from coding errors, but from design problems, requirement oversights, platform or component incompatibilities, unexpected data, timing dependencies… Bugs may appear in a system even though the units or components have been tested and seem to be just fine on their own. Problems happen due to unforeseen conditions, interactions, interruptions, or inadvertent misuse. That is, many bugs are <em>emergent</em>.</p><p>Another real possibility is that the generated code will omit features, or things like error checking and exception handling, even if you prompt for them. Unlike you, the LLM won’t ponder the request before it starts typing. The LLM will start spitting output immediately. An LLM doesn’t reflect on the ideas it has, because it doesn’t doesn’t have ideas; it processes tokens. For a sufficiently complex product, the LLM may run out of tokens in its context window, rendering it forgetful about earlier prompts or responses, or code generated earlier in the session.</p><p>Remember, the LLM doesn’t understand things; it calculates probable next words or strings. It doesn’t deal at all with the essential human parts of software development. It doesn’t socialize, manage politics, refine requirements, or negotiate the mission in the ways you’ve learned to do. It doesn’t respond to a puzzled look.</p><p>As a tester, I’m naturally concerned about the possibility of problems, bugs, and risks about the product we’re developing. I’m definitely concerned that LLMs aren’t as technically competent as you are. After all, even though they have access to more data and more programs than we could ever have, they regularly screw up at simple tasks associated with <a href="https://developsense.com/blog/2023/11/bing-chat-the-evaluate-function-and-the-wolfram-alpha-plugin">math</a>, <a href="https://developsense.com/blog/2023/11/chatgpt-and-test-data">data generation</a>, <a href="https://developsense.com/blog/2023/08/experience-report-using-chatgpt-to-generate-and-analyze-text">text analysis</a>, <a href="https://developsense.com/blog/2023/08/chatgpt-and-a-math-puzzle">reasoning about puzzles</a>, and even <a href="https://developsense.com/blog/2023/09/reliably-unreliable">describing themselves</a>.</p><p>Since it doesn’t participate in human society, an LLM doesn’t have your social competence either. People make stereotyped jokes about programmers. (Q. How many programmers does it take to change a light bulb? A. Let me tell you why it’s not a good idea to change the light bulb…) But programmers act like programmers because human beings have skills and tacit knowledge that affect their missions. An LLM doesn’t have a conscience, or empathy. It doesn’t deal with controversy between stakeholders. It doesn’t obtain real human experience with your product — or with <em>any</em> product — so it can’t recognize many things that real people might like or dislike about it.</p><p>Some people who tout LLMs say that they’re helpful with test design. I hope for everyone’s sake that LLMs are better at coding than they are at testing tasks, because they’re <a href="https://www.satisfice.com/download/appendix-chatgpt-sucks-at-being-a-testing-expert">definitely not very good at those</a>. (More on that <a href="https://developsense.com/blog/2023/10/a-reply-to-running-a-crowd-sourced-experiment-on-using-llms-for-testing">here</a>, and <a href="https://developsense.com/blog/2023/10/a-reply-to-running-a-crowd-sourced-experiment-on-using-llms-for-testing-part-2-analysis">here</a> too.) That’s explicable; LLMs are trained on the internet, where there’s a stonking amount of bad writing about testing, and even worse advice about it.</p><p>(My colleague <a href="http://www.satisfice.com/">James Bach</a> points out that while there has been enormous effort and some success associated with suppressing prejudice, hate speech, racism, and sexism, in LLM output, there’s apparently been no push to reduce the volume of bad testing advice.)</p><p>Since LLMs are not technically savvy, nor social agents, nor socially competent, it would be a mistake to trust LLMs for most kinds of testing tasks today, or at any time in the foreseeable future. They <a href="https://www.linkedin.com/feed/update/urn:li:activity:7131301704758677505/">might be helpful when there’s not very much at stake</a>, and when the tester remains in charge, but any output that matters from an LLM will need scrutiny for a good while yet. Moreover, the volume of stuff being produced can be overwhelming. Reviewing LLM output carefully is like looking for unexploded bombs after an avalanche.</p><p>Now, there are some skilled testers who are ready and eager right now to help with testing of LLMs — or at least getting ready to help. It’s true; there aren’t nearly enough of us yet. We’re going to need lots more, and we’re all going to have to practice and talk about what we’re learning. The overarching theme of testing — getting experience with products and analyzing them with a focus on spotting trouble before it’s too late — doesn’t change. In <a href="https://www.rapid-software-testing.com/">Rapid Software Testing</a>, at least, we’re always working on ideas on how to help people test new things, think critically about them, and analyze risk. We’re super happy to work with developers on that, too.</p><p>Meanwhile, here’s a list of things I don’t hear my developer friends saying very often:</p><p>“I LOVE testing!”<br>“I can’t wait to find hidden, rare, subtle, intermittent, emergent bugs in this product!”<br>“Reading through piles of unreliable code is great!”<br>“Putting ths thing through its paces to find deep problems? When can I start?!”<br>“There’s a possibility that this thing might spill people’s confidential data; I’m goin’ a-huntin’ for security holes!”<br>“Fixing bugs in code written by others is my idea of a good time!”<br>“Supervising junior programmers that don’t learn from experience is fun!”</p><p>And this is what I’m warning you about, Dear Developer: development processes and tools that incorporate Large Language Models will produce enormous piles of code very quickly. You won’t be writing that code, but if you’re responsible for it, you’ll be responsible for examining, reviewing it, finding problems in it, and fixing it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/1*esw7gzTDZx5pj3qDd1AVyw.jpeg" /><figcaption>“This guy still isn’t getting it. Why is he still not getting it?”</figcaption></figure><p>That is: if you are using LLMs, or if you’re directed to do so, you’re about to turn from an author of your own code into a reviewer and maintenance programmer — and maybe a tester, too — for a barely competent programmer producing mountains of mediocre and unreliable code.</p><p>Are you ready for that?</p><p>Your friend,</p><p>Michael B.<br>Tester<br><a href="https://developsense.com/">https://developsense.com</a></p><p><em>I’d love to receive your reply on how you think about this. Plus I was reading a piece of advice today, suggesting that I’m supposed to tell you to hit that Clap or Like or Heart button so that more people see this post. So if you do that, thanks.</em></p><p><em>Original post at </em><a href="https://developsense.com/blog/2023/11/to-the-developer-about-your-impending-promotion"><em>https://developsense.com/blog/2023/11/to-the-developer-about-your-impending-promotion</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=86f27ac99025" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>