<?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[VRONT - Medium]]></title>
        <description><![CDATA[Everything not saved will be lost - Medium]]></description>
        <link>https://medium.com/thymos-x-mimesis?source=rss----ddd852b9eaa8---4</link>
        <image>
            <url>https://cdn-images-1.medium.com/proxy/1*TGH72Nnw24QL3iV9IOm4VA.png</url>
            <title>VRONT - Medium</title>
            <link>https://medium.com/thymos-x-mimesis?source=rss----ddd852b9eaa8---4</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 13 Nov 2026 02:36:46 GMT</lastBuildDate>
        <atom:link href="https://medium.com/feed/thymos-x-mimesis" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Say hello to Apple's next CEO]]></title>
            <link>https://medium.com/thymos-x-mimesis/say-hello-to-apples-next-ceo-74d83e0862ab?source=rss----ddd852b9eaa8---4</link>
            <guid isPermaLink="false">https://medium.com/p/74d83e0862ab</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[apple]]></category>
            <category><![CDATA[agentic]]></category>
            <dc:creator><![CDATA[Matthias "Mattes" Schrader]]></dc:creator>
            <pubDate>Wed, 12 Mar 2025 14:19:30 GMT</pubDate>
            <atom:updated>2025-03-13T03:01:43.145Z</atom:updated>
            <content:encoded><![CDATA[<h3>Yo Say hello to Apple&#39;s next CEO</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*M6_zvLbam7-X7SCtRKNbGQ@2x.jpeg" /></figure><h3><strong>I just saw Apple’s future CEO at SXSW: Meredith Whittaker might finally be the one to catapult Cupertino back into the AI spotlight.</strong></h3><p>Right now, Apple is neck-deep in a quagmire. Since last summer, they’ve been drumming up hype about “Apple Intelligence,” but all Siri’s delivering irrelevance instead of inference: The product is a fiasco. TV ads? Pulled. Launch date? Kicked down the road. Tim Cook? Rudderless (and not just since his bizarre Trump donation).</p><p>Here’s the absurd part: The rise of LLMs handed Apple a golden opportunity. Exactly two years ago, <a href="https://medium.com/thymos-x-mimesis/chatgpt-4-may-be-the-best-thing-that-could-happen-to-apple-3432fb33ba12">I wrote here on Medium</a> that LLMs could help Apple reinvent the idea of the personal computer for the fourth time. Not as some all-knowing chatbot, but as a hyper-individualized digital alter ego — something only Apple could pull off with its trinity of user trust, privacy DNA, and top-to-bottom control of hardware and software. They could’ve transformed their ecosystem into an invincible “egosystem.”</p><p>Fifteen months later, Apple finally rolled out that very concept at the latest WWDC. But judging by the underwhelming demo — like the painfully basic “What time is my mom’s flight landing?” — Siri is about as close to real AI wizardry as Tesla is to being a cuddly Love Brand.</p><p>At SXSW, Whittaker <a href="https://www.youtube.com/watch?v=AyH7zoP-JOg">delivered a laser-precise autopsy</a> of why Apple’s failing here — without even naming the company:</p><blockquote>“Agentic systems have to break the blood/brain barrier of the App/OS Layer. To do all the cool things they need root permissions and there is no model to do this encrypted.”</blockquote><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FAyH7zoP-JOg%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DAyH7zoP-JOg&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FAyH7zoP-JOg%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/5bc5463ffa7b37f133c068fc2509c1cf/href">https://medium.com/media/5bc5463ffa7b37f133c068fc2509c1cf/href</a></iframe><p>Translation from Whittaker-speak: A beneficial AI has to tap deep into a user’s data graph — contacts, calendars, preferences, digital identity, and all the other stuff you wouldn’t share with a soul. Never. But that’s the crux: LLMs aren’t precisely bulletproof with facts nor safe from manipulations like prompt injection. In plain English, they can be tricked into doing things they absolutely shouldn’t do.</p><p>Now Apple feels the brutal consequences of treating LLMs like a plug-and-play commodity and betting on OpenAI and Google Gemini instead of charting its own path. Anyone serious about baking AI into the OS at a deep level needs a robust, proprietary model that puts privacy front and center. So it’s downright absurd that Apple has been pushing “Apple Intelligence” for a year even though it’s nowhere near ready for primetime — and conceptually stuck in a dead end.</p><p>So, would I be surprised if Apple veteran Guy Kawasaki actually interviewed the next Apple CEO on that stage? Not at all. Cupertino desperately needs someone like Meredith Whittaker to slam the brakes on the recent years of misfires (hello, Apple Vision and Apple Car?) and bring the trillion-dollar giant back to relevance with some clear-eyed leadership.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=74d83e0862ab" width="1" height="1" alt=""><hr><p><a href="https://medium.com/thymos-x-mimesis/say-hello-to-apples-next-ceo-74d83e0862ab">Say hello to Apple&#39;s next CEO</a> was originally published in <a href="https://medium.com/thymos-x-mimesis">VRONT</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Kulturrevolution 2.0 — Made in Hangzhou]]></title>
            <link>https://medium.com/thymos-x-mimesis/kulturrevolution-2-0-made-in-hangzhou-66eeed846013?source=rss----ddd852b9eaa8---4</link>
            <guid isPermaLink="false">https://medium.com/p/66eeed846013</guid>
            <category><![CDATA[deepseek]]></category>
            <category><![CDATA[open-source]]></category>
            <category><![CDATA[kultur]]></category>
            <dc:creator><![CDATA[Matthias "Mattes" Schrader]]></dc:creator>
            <pubDate>Thu, 30 Jan 2025 10:26:22 GMT</pubDate>
            <atom:updated>2025-01-30T19:32:21.633Z</atom:updated>
            <content:encoded><![CDATA[<h3><strong>Kulturrevolution 2.0 — Made in Hangzhou</strong></h3><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2Fd0pKXozR3Ek%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3Dd0pKXozR3Ek&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2Fd0pKXozR3Ek%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="640" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/c73c0b86fb7a1b7868973595b33aefb3/href">https://medium.com/media/c73c0b86fb7a1b7868973595b33aefb3/href</a></iframe><p><strong>Der Januar endet mit einem absurden Plot-Twist: Während Donald Trump mit seinen neuen Buddies aus dem Silicon Valley Hunderte von Milliarden Dollar für eine KI-Initiative verkündet, die eher nach </strong><a href="https://youtu.be/b17q_fpa90o?si=12bi6QQBuB4YjMYc"><strong>B-Movie Science-Fiction</strong></a><strong> klingt, und gleichzeitig Microsoft neue Datacenter, so groß wie Manhattan plant, stellt das Sidequest eines Hedge-Funds aus Hangzhou die KI-Welt auf den Kopf. Hier formt sich etwas Neues — still, radikal, fast schon romantisch. Deepseek, das sich mehr als Forscherkollektiv denn als Startup begreift, stellt die Grundfesten des Tech-Kapitalismus infrage. Und exakt hier liegt eine Blaupause für Deutschland.</strong></p><p>Deepseek geht es im Kern um die Enträtselung menschlicher Intelligenz und Bewusstseins durch künstliche Nachbildung (AGI). Das Team verbindet diese Vision mit einem Open-Source-Idealismus und einer Kultur, die an die früheren Hackerspaces erinnert. “<a href="https://media.ccc.de/c/27c3">We come in Peace</a>” schrieb seinerzeit der Hamburger Chaos Computer Club. Deepseeks CEO Liang Wenfeng, ein eher schüchterner Technik-Nerd aus der Provinz Guangdong, <a href="https://thechinaacademy.org/interview-with-deepseek-founder-were-done-following-its-time-to-lead/">formuliert es so</a>:</p><blockquote>„Die Bereitstellung von Cloud-Diensten ist nicht unser Hauptziel — das Erreichen von AGI ist es. Das erfordert Neugier und kreativen Ergeiz. Wir sind noch gefesselt an der Business-Getriebenheit der alten Zeit“.</blockquote><p>Dies ist ein zutiefst radikaler, ja romantischer, Ansatz, der so aus der Zeit gefallen klingt, wie er nur sein kann.</p><h4>Deepseek ist Deep Tech</h4><p>Während das Silicon Valley auf „brute force“-Produktionen setzt — mehr Funding, mehr GPUs, mehr User –, betreibt Deepseek ein Indie-Studio mit einer Liebeserklärung an das Unbekannte. Während OpenAI als Closed Shop alle Details zu GPT-5 geheim hält (in Davos machten sogar Gerüchte über Probleme bei der Entwicklung die Runde), publiziert Deepseek seine wissenschaftlichen Durchbrüche <a href="https://arxiv.org/search/?query=deepseek&amp;searchtype=all">detailliert auf arXiv</a> und teilt die Modelle als Open Source.</p><p>Die neuen Large Language Models (LLMs) von der chinesischen Ostküste rekalibrieren die Preise im globalen KI-Markt auf Temu-Level. Nicht, wie von einigen schnell auf X gemutmaßt, durch einen Bootleg-Remix der bestehenden Frontier-Modelle, sondern durch architektonische Innovationen. Das sogenannte Multi-Head Latent Attention (MLA) schrumpft massiv den Speicherbedarf herkömmlicher LLMs — ein Quantensprung. Eine weitere Neuerung ist der Mixture-of-Experts (MoE) genannte Ansatz: Bekommt das zugrunde liegende LLM eine Nutzeranfrage, wird nicht immer das ganze Modell aktiviert, sondern nur die Teile, die zur Beantwortung dieser Frage gebraucht werden. OpenAI-Veteran Miles Brundage kommentiert diese Konzepte ehrfürchtig; sie seien „voll erstaunlicher Klugheit“ — übersetzt: “Shit, die Chinesen haben uns in unserem eigenen Game gecheated”.</p><h4><strong>Hackerkultur</strong></h4><p>Deepseeks genialster Hack ist jedoch: Sie verstehen Open Source nicht als Distributionstaktik, sondern als kulturellen Akt. Durch das Teilen des Codes zwingt das Unternehmen den Rest der Welt zum Mitziehen und schafft ein Ökosystem, das von seinem Know-how abhängt. Über Nacht sind weltweit Dutzende Cloud-Anbieter auf den Zug aufgesprungen und bieten den Zugriff auf Deepseeks Modelle zum Kampfpreis an — mit nichtzensierten Trainingsdaten und ohne Datentransfer nach China.</p><blockquote>“Daher liegt unser wirklicher Burggraben im Wachstum unseres Teams — dem Aufbau von Know-how und der Förderung einer innovativen Kultur. Open Source und die Veröffentlichung von wissenschaftlichen Papers führen nicht zu nennenswerten Verlusten. Für und als Techies ist es lohnend, nachgeahmt zu werden. Open Source ist kulturell, nicht nur kommerziell. Etwas zurückzugeben ist eine Ehre und zieht Talente an.”</blockquote><p>„Unser Burggraben ist unser Team“, sagt Liang Wenfeng — eine Haltung, die an die Linux-Ära der 2000er erinnert, als Open Source die damaligen Giganten wie Sun Microsystems oder Silicon Graphics ausradierte.</p><p>Anthropic-Mitgründer Jack Clark spricht über Deepseeks Team von einer Generation von 20-Jährigen, die „unfassbare Genies“ seien. Deren Gründer Liang Wenfeng hängt es wesentlich tiefer: „Das sind einfach junge Absolventen, die Lust auf harte Probleme haben.“ Ohne Hierarchien, ohne feste Rollen, ohne Budget-Bullshit. Wer eine Idee hat, bekommt Zugang zu Rechenleistung, ohne Antrag. Diese intrinsische Lust am Machen ist das Gegenteil der heutigen Westküsten-Broscience-Kultur, wo Talente mit obszönen Aktienoptionen geködert werden. Stattdessen wirkt Deepseek wie der Homebrew Computer Club in Menlo Park der 1970er Jahre, wo Steve Jobs und Steve Wozniak ihre Leidenschaft für Elektronik entdeckten und den ersten Personal Computer bastelten.</p><h4><strong>Von der Werkbank zum Denkwerk der Welt</strong></h4><p>China wandelt sich von der Werkbank zum Denkwerk der Welt. Auch deswegen sollte uns Deepseek in Europa und insbesondere in Deutschland maximal nerven: Hätten wir das nicht auch erreichen können? Die zentrale Botschaft für uns lautet: Biss statt Budget bewegt die Welt. Wir haben uns zu lange eingeredet, gegen die großen Technologiewetten der Kapitalsammelstellen an der Sand Hill Road keine Chance zu haben. Wir haben uns in der komfortablen Zeitblase vergangener Industrien eingerichtet, uns auf Innovationsdiät gesetzt und unsere Ambitionen herunterkalibriert. Wir brauchen eine Rückbesinnung auf langfristiges, neugiergetriebenes Handeln — fernab von Börsenkursen und Pitch Decks. Wir sind reich an Forschungsinstituten, reich an Talenten, reich an Unternehmen. Aber unsere Tech-Kultur ist geizig an Ambition und Mut.</p><p>Deutschland hat genügend kluge Köpfe — doch viele wandern ab, oder verkümmern in steifen Corporates und Startups mit ihrem profanen „Growth Hacking“-Kult. Was fehlt, ist eine Kultur, die Scheitern und Verschwendung erlaubt, ambitioniertes Spiel ermöglicht, technologische Abenteuer belohnt und — vor allem — sich traut, transzendente Sinnstiftung zu vermitteln. Was könnte es Größeres geben, als daran zu arbeiten, menschliche Intelligenz und Bewusstsein zu artifizieren? In diesem Sinne ist Deepseek auch ein Manifest. Es beweist: Tech-Geschichte kann auch off-road der Sand Hill Road geschrieben werden. Ein Plädoyer für eine Kultur, in der Neugier, Technologie-Romantik und die Liebe zum Gelingen die Triebkräfte sind — und nicht die nächste Funding-Runde oder das nächste Ethik-PDF für den Projektantrag. Oder um Liang Wenfengs Credo zu zitieren:</p><blockquote>„Ich glaube, dass Innovation in erster Linie eine Frage des Glaubens ist. Warum ist das Silicon Valley so innovativ? Weil sie es wagen, es zu versuchen“.</blockquote><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=66eeed846013" width="1" height="1" alt=""><hr><p><a href="https://medium.com/thymos-x-mimesis/kulturrevolution-2-0-made-in-hangzhou-66eeed846013">Kulturrevolution 2.0 — Made in Hangzhou</a> was originally published in <a href="https://medium.com/thymos-x-mimesis">VRONT</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Prometheus Prompt]]></title>
            <link>https://medium.com/thymos-x-mimesis/the-prometheus-prompt-ff9d5b086e21?source=rss----ddd852b9eaa8---4</link>
            <guid isPermaLink="false">https://medium.com/p/ff9d5b086e21</guid>
            <category><![CDATA[openai]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[gpt-4]]></category>
            <dc:creator><![CDATA[Matthias "Mattes" Schrader]]></dc:creator>
            <pubDate>Wed, 29 Jan 2025 10:12:55 GMT</pubDate>
            <atom:updated>2023-04-03T10:01:23.006Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="Midjourney: marble statue of Prometheus with a keyboard, shooting for a magazine cover, professional light — ar 16:9 — v5" src="https://cdn-images-1.medium.com/max/1024/1*fpSWn3eqUohYUHgGacC4zw.jpeg" /></figure><h4><strong>The fear about generative AI systems comes not from their power but from their similarity to our own inner construction of reality: modeling, prediction, and hallucination with recursive training loops. It is the fear of ourselves and our non-perfection. And the awe of further slights.</strong></h4><blockquote>Prometheus:<br>Ay, but for me, mankind had never had<br>The art of fire, that great resource, but kept<br>By Gods alone in their own citadel:<br>And this it was I filched for mortal men.</blockquote><blockquote><em>— Aeschylus. “Prometheus Bound”</em></blockquote><p><strong>In Greek mythology, Prometheus symbolized the rebellious spirit and our relentless purpose for progress. The divine fire represents our human ability to create tools and technologies that improve our lives and shape the environment. It represents the transition from a raw, animal-like existence to a higher order where humans can control and shape their environment. The ultimate tool homo sapiens have created for this purpose — or was given to us by the gods if we follow the myths of the ancient Greeks — is language encoded as text.</strong></p><p>Andy Clark is the most quoted philosopher of our time. In his new and breathtaking book “<a href="https://www.penguin.co.uk/books/313594/the-experience-machine-by-clark-andy/9780241394526">The Experience Machine</a>,” he describes our brain does not perceive the world in the first instance through our senses. Still, our mind is a prediction machine, which is only constantly trained with the input recorded through our senses. Therefore, we live in a simulation universe, which is located between our skull bones. We strive to shape the external world to our internal simulation to minimize the error-handling routines of our inner experience machine if the outer environment is not in sync with our imagination.</p><p>If this were so, the occasional shivering at the capabilities of generative AI systems is <a href="https://every.to/chain-of-thought/gpt-4-is-a-reasoning-engine">explained by their similarity</a> to our own construction of reality: modeling, prediction, and hallucination with recursive training loops. In this sense, it is in some ways understandable that this chill sometimes turns into a shiver, resulting in petitions like last week’s <a href="https://futureoflife.org/open-letter/pause-giant-ai-experiments/">Open Letter from the Future of Life Institute</a>:</p><blockquote>AI systems with human-competitive intelligence can pose profound risks to society and humanity, as shown by extensive research and acknowledged by top AI labs… Therefore, we call on all AI labs to immediately pause for at least six months the training of AI systems more powerful than GPT-4.</blockquote><p>However, I believe the first signatories are more interested in the attention and satisfaction of their narcissism. GenAI has all the attributes to add a fourth fundamental slight to humanity after the Copernican turn, Darwin’s evolution theory, and Freud’s psychoanalysis: Intelligence is not bound to human hosts, and creativity is nothing else than the ability to exude good filters for hallucinations.</p><h4>Prometheus Prompt is in the world</h4><p>Behind the Open Letter and many approving comments, some of which make <a href="https://time.com/6266923/ai-eliezer-yudkowsky-open-letter-not-enough/">much more radical demands</a>, is the idea that technological progress can be stopped, like in the case of the construction of atomic bombs, where states have agreed on non-use and non-proliferation. And where this arrangement is circumvented, it still takes state actors to enrich weapons-grade uranium to an appropriately high level. But this is entirely different regarding GenAI. The signatories, most of whom come from the tech and AI scene, know this too. The Prometheus prompt is out in the world and here to stay. Let’s walk through the building blocks of an LLM and see why.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Hnd5iy3XE0GxNyBoGVMpug.jpeg" /></figure><h4>Cluster &amp; Cloud</h4><p>The training of ChatGPT-3 required 1024 GPUs, took 34 days, and<a href="https://arxiv.org/abs/2104.04473"> cost $4.6M in computing alone</a>. ChatGPT-4 is orders of magnitude larger, but the <a href="https://medium.com/codex/gpt-4-will-be-500x-smaller-than-people-think-here-is-why-3556816f8ff2">rapid optimization of algorithms</a> and <a href="https://nvidianews.nvidia.com/news/nvidia-hopper-gpus-expand-reach-as-demand-for-ai-grows?nvid=nv-int-csfg-738120#cid=gtcs23_nv-int-csfg_en-us">hardware</a> partially mitigates the increased complexity. Even if the next generation of LLMs, such as with GPT-5, has further growing infrastructure requirements, this does not necessarily require Big Tech. The crypto-mining community provides the blueprint. Similarly, GPU performance is what counts most when training generative AI systems. There are no central instances in mining but the primacy of decentralization. The first principles of how next-generation AI models can be computed decentralized <a href="https://www.together.xyz/blog/neurips-2022-overcoming-communication-bottlenecks-for-decentralized-training-2">already exist</a>. The required TFlops of GPU performance are provided by Apple with about 30 million Macs with Apple MX Silicon and NVIDIA’s shipments of GPU-heavy graphics cards this year. In the coming months, we will see very good LLMs computed peer-to-peer on large, decentralized GPU clusters. In parallel, NVIDIA will use the tailwind of the AI hype to sell its new GPU generation to data center &amp; cloud providers for the hell of it to maintain its previous success story (<a href="https://www.alphaspread.com/security/nasdaq/nvda/summary">and high market valuation</a>).</p><h4>Copy Code</h4><blockquote>“This is the Unix philosophy: write programs that do one thing and do it well. Write programs to work together. Write programs that handle text streams, because that is a universal interface.” —Douglas McIlroy.</blockquote><p>With <a href="https://en.wikipedia.org/wiki/Unix_philosophy">the Unix Manifesto</a> from 1978, Ken Thompson and Doug McIlroy have had a powerful influence on the software development culture until today. Very many concepts from then are more relevant today than ever:</p><ul><li>Composable Applications</li><li>Open source, which is easily understandable and changeable</li><li>Interoperability of applications through open APIs</li><li>Text as a universal interface</li></ul><p>These principles were the secret source that propelled the Internet as a decentralized network to become a global success. The servers of SUN (Stanford University Network), which made up the Internet’s backbone, and the workstations of Steve Jobs’s NeXT, on which the first web pages were created in CERN, ran on Unix. Almost everything we use as software today can be rooted in the above principles. The culture of open source also applies to the Large Language Models of generative AI. There are now numerous alternatives to OpenAI’s models on GitHub (such as <a href="https://github.com/nomic-ai/gpt4all">gpt4all</a>), and a fast-growing developer community is pouncing on them. An anecdote is that LLMs are particularly good at understanding programming languages and generating code. LLMs introduce whole cohorts of newbies to software development, some of whom improve the models again. LLMs feed LLMs.</p><h4>Content Commodity</h4><p>With the triumph of the Web, text became the universal interface for sharing the world’s knowledge in the spirit of the Unix founders. HTML is essentially a text markup language — thus readable, referenceable, and weighable. With Google’s almost monopolistic success, an entire industry emerged with SEO, which ensured that all content was optimally prepared for search engine crawlers.</p><p><a href="https://developers.google.com/search/docs/fundamentals/seo-starter-guide">With its SEO rules</a>, Google built the API for the world knowledge of humankind. Projects like <a href="https://commoncrawl.org/the-data/get-started/">Common Crawl</a> take advantage of this, crawl around 1.5 billion websites per month, and make the datasets available for free (including the last ten years). <a href="https://openwebtext2.readthedocs.io/en/latest/background/">OpenWebText2</a> supplements this data with additional scrapes and weights from social networks (like Reddit).</p><p>All the text on the Web can now be stored on a stack of hard drives that cost less than <a href="https://en.wikipedia.org/wiki/Macintosh_128K">the first Macintosh 128k</a> when it was launched in 1984.</p><h4>Transformer Architectures</h4><p>So the essential building blocks for GenAI and LLMs today are a commodity: content, code, and cloud/cluster-based GPU infrastructures. That’s also true for the last part of the puzzle. The crucial breakthrough in AI came with the idea of the so-called Transformer architectures from a team of researchers at Stanford in 2017 (<a href="https://arxiv.org/pdf/1508.04025.pdf">paper</a>, <a href="https://nlp.stanford.edu/projects/nmt/">code</a>). The Transformer’s mathematics of squeezing gigantic amounts of text into weighted matrices and enriching them with clever, attention-grabbing mechanisms enabled LLMs, for the first time, to process large amounts of textual data and learn to understand the meaning of language at a deeper level. With the Transformer architecture, the development of LLMs in their current form and performance was possible. The mathematics and algorithms behind it are now part of the <a href="https://web.stanford.edu/class/cs25/">curriculum in computer science</a>.</p><h4>Wrap-up</h4><p>ChatGPT is probably the most <a href="https://www.ubs.com/global/en/wealth-management/our-approach/marketnews/article.1585717.html">successful product launch ever</a>. People immediately see the added value in the product and rather quibble with the fact that ChatGPT can (still) do too little than it is too powerful.</p><p>The entire stack for building (decentralized) LLMs is open and freely available. The British philosopher Andy Clark explains that our brain is a constant prediction machine, trained with the input picked up by our senses. If the resonant loops of modeling, prediction, and hallucination with sensory feedback are the core of intelligence, then the senses of LLMs are the prompts and input of humans. LLMs train themselves through and on us to get better.</p><p>We all know the formula for freewheeling platforms:</p><ol><li>initial disruptive product (here: ChatGTP)</li><li>pull of enthusiastic users</li><li>collecting feedback as training data</li><li>training of LLM through feedback</li><li>better product experience (through training data)</li><li>pull from more enthusiastic users</li><li>goto 3</li></ol><p>In our open and competitive societies and markets, the notion of a central moratorium is therefore naïve or even ridiculous, to quote media professor and book author (“What would Google do?”) Jeff Jarvis.</p><h3>Jeff Jarvis on Twitter: &quot;Oh, this is absolutely fucking ludicrous and laughable, a prime specimen of #MoralPanic if I&#39;ve ever seen one: A call to pause training AI. Gutenberg: Stop your presses! Signed by so many usual-suspect moral entrepeneurs &amp; attention addicts. https://t.co/e92ZKWVWik / Twitter&quot;</h3><p>Oh, this is absolutely fucking ludicrous and laughable, a prime specimen of #MoralPanic if I&#39;ve ever seen one: A call to pause training AI. Gutenberg: Stop your presses! Signed by so many usual-suspect moral entrepeneurs &amp; attention addicts. https://t.co/e92ZKWVWik</p><p>The prompt of the good old Bash shell was not a “$” for nothing. As humans, we have never been able to resist the promises of better tools. Our desire to shape the world according to our hallucinations and thoughts with better tools was consistently more potent in the end. And finally, we must have the courage to withstand the potential slights of GenAI. The non-perfection of AI is a painful mirror of ourselves, and it is the price we have to pay Godfather Zeus for our rebellion.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ff9d5b086e21" width="1" height="1" alt=""><hr><p><a href="https://medium.com/thymos-x-mimesis/the-prometheus-prompt-ff9d5b086e21">The Prometheus Prompt</a> was originally published in <a href="https://medium.com/thymos-x-mimesis">VRONT</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[I built my very first Personal Chatbot powered by OpenAI and my Mac mini, Here is How]]></title>
            <link>https://medium.com/thymos-x-mimesis/how-i-built-my-very-first-personal-chatbot-appliance-pca-powered-by-chatgpt-and-my-mac-mini-f6aeeedc7654?source=rss----ddd852b9eaa8---4</link>
            <guid isPermaLink="false">https://medium.com/p/f6aeeedc7654</guid>
            <category><![CDATA[gtp]]></category>
            <category><![CDATA[chatbots]]></category>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[gpt-3]]></category>
            <category><![CDATA[openai]]></category>
            <dc:creator><![CDATA[Matthias "Mattes" Schrader]]></dc:creator>
            <pubDate>Wed, 29 Jan 2025 10:12:35 GMT</pubDate>
            <atom:updated>2023-03-30T08:53:57.123Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*cVSAapnLw5GHTmaxTNegjg.jpeg" /></figure><ul><li><strong>A Step-by-step guide on how you can build your local chatbot</strong></li><li><strong>Private-infused LLM on your own computer and control over your data</strong></li><li><strong>Achtung! the OpenAI API is not a bargain: 1GB eats $1,000</strong></li></ul><p>OpenAI has fully embraced Google’s SEO-curated web and transformed it with ChatGPT into something, well, new. Some are convinced to see here <a href="https://www.microsoft.com/en-us/%20research/publication/sparks-of-artificial-general-intelligence-early-experiments-with-gpt-4/">the first sparks of general intelligence</a>. However, with the <a href="https://openai.com/blog/chatgpt-plugin">new ChatGPT plugin architecture</a> released last week, OpenAI follows the classic platform playbook of the software industry: flip competing products into features of existing products. Countless startups are plugged out by the plugins overnight.</p><p>But — <a href="https://medium.com/@mattesmattes/chatgpt-4-may-be-the-best-thing-that-could-happen-to-apple-3432fb33ba12">I still miss the chance of carving out an on-device Chatbot</a> that intimately interacts with all my concepts, presentations, contracts, account statements, and emails from the last few decades. I cannot deny that this desire is maybe only an echo from the past; been in this industry too long. Before the Adwords money printing machine ran at full velocity, Google tried to find the product/market fit with their Google Search Appliance (GSA). Back then, the idea was super simple: transform internal data deserts of enterprise customers with Google’s <a href="https://en.wikipedia.org/wiki%20/Google_Search_Appliance#Version">on-prem yellow pizza boxes</a> into a flourishing oasis of insights for managers. I’m shamelessly copying the GSA history here to name my little LLM hobby project a Personal Chatbot Appliance (PCA).</p><p>I know startups like <a href="https://www.chatbase.co/">Chatbase</a> or <a href="https://meetcody.ai/">Cody</a> are trying to fill this gap. Admittedly, quickly setting up and sharing your chatbot there is highly convenient. But the scope is more still a Limited Language Model. Chatbase only accepts around 2 MB of text, and I also failed with larger files with Cody. And finally: who trusts their entire life dump to vulnerable startups?</p><p>Having a PCA on my desk would be cool. So, let’s go!</p><p>This step-by-step guide has three parts:</p><ul><li>set up the development environment,</li><li>prepare your data and</li><li>build &amp; run the app finally.</li></ul><h4>1. Development Environment</h4><p>A few things must first be installed on your computer for data preparation and running your chatbot. The procedure for all systems (Mac, Windows, Linux) is very similar. In the following, I will only go into the Mac. If there are problems or error messages during setup, use ChatGPT for help. It saved me hours!</p><p>First, open a terminal window and install Homebrew as a package manager, which MacOS doesn’t have.</p><pre>/bin/bash -c &quot;$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)&quot;</pre><p>At the end of the homebrew installation, please follow the instructions in the terminal window to define the environment variables for the paths. We can then install Python (Python3 on a Mac):</p><pre>brew install python</pre><p>Then install PIP as a new package manager:</p><pre>python3 -m pip install -U pip</pre><p>Also here the environment variables must be set correctly again:</p><pre>echo ‘export PATH=”/Users/mattes/Library/Python/3.9/bin:$PATH”’ &gt;&gt; ~/.bash_profile</pre><pre>source ~/.bash_profile</pre><p>Now we are ready to install the critical OpenAI libraries:</p><pre>pip install openai</pre><p>Next, let’s install GPTIndex, which is also called LlamaIndex. It allows the LLM to connect to the external data that is our knowledge base:</p><pre>pip install gpt_index</pre><p>The following libraries are required for data preparation and conversion:</p><pre>pip install mailbox</pre><pre>pip install unicode </pre><pre>pip install html2text</pre><pre>pip install unidecode</pre><p>Finally, I installed Gradio as a web GUI for the chatbot:</p><pre>pip install gradio</pre><p>If you still need to install a code editor, I recommend<a href="https://apps.apple.com/de/app/bbedit/id404009241?mt=12"> BBEdit on the Mac</a>.</p><h4>2. Data Preparation</h4><p>My Gmail inbox is my life cache. So I intend to train my PCA with emails and all the other staff from my SSDs. Because this is the promise of transformational architectures: dump whole piles of barely filtered, mostly unedited scrapes of your data footprint into the eager maw of a hungry model.</p><p>So much for the theory; using the OpenAI API is expensive in practice. It is charged in tokens, which is equivalent to one word in the English language. 1000 tokens cost $0.002 — or about $1 per MB. That looks like little at first glance. However, my mailbox contains about 200 GB, and the local SSD has another 800 GB of data (without photos and videos). So one TB of training data would cost no less than one million dollars (so protect your private API key!).</p><p>When processing your training data, you need a strategy. I decided to set a limit of $100 or around 100 MB for myself. I have set aside about 50 MB of data from my mailbox and another 50 MB for various documents.</p><p>I just exported the <strong>sent</strong> folder with <a href="https://takeout.google.com/">Google Takeout</a> to reduce the data from the mailbox as a first step. Then I converted the downloaded <strong>.mbox</strong> archive to plain <strong>.txt</strong> with the following script (plus it delete all attachments and markups):</p><pre>import os<br>import mailbox<br>import email<br>from email.header import decode_header<br><br>def get_subject(msg):<br>    raw_subject = msg.get(&#39;Subject&#39;)<br>    if not raw_subject:<br>        return &#39;&#39;<br><br>    subject = decode_header(raw_subject)<br>    decoded_subject = &#39;&#39;<br>    for s, encoding in subject:<br>        if encoding:<br>            try:<br>                decoded_s = s.decode(encoding)<br>            except LookupError:<br>                # handle unknown encoding<br>                decoded_s = s.decode(&#39;utf-8&#39;, &#39;ignore&#39;)<br>            decoded_subject += decoded_s<br>        else:<br>            decoded_subject += s if isinstance(s, str) else s.decode(&#39;utf-8&#39;, &#39;ignore&#39;)<br>    return decoded_subject<br><br>def get_body(msg):<br>    def get_decoded_payload(part):<br>        payload = part.get_payload(decode=True)<br>        charset = part.get_content_charset() or &#39;utf-8&#39;<br>        try:<br>            return payload.decode(charset)<br>        except LookupError:<br>            # Use a fallback encoding value if the charset is not recognized<br>            return payload.decode(&#39;utf-8&#39;, &#39;ignore&#39;)<br>        except UnicodeDecodeError:<br>            # Handle UnicodeDecodeError if the payload cannot be decoded<br>            return payload.decode(&#39;utf-8&#39;, &#39;ignore&#39;)<br><br>    if msg.is_multipart():<br>        for part in msg.walk():<br>            if part.get_content_type() == &#39;text/plain&#39;:<br>                return get_decoded_payload(part)<br>    else:<br>        return get_decoded_payload(msg)<br><br>def export_emails_to_txt(mbox_path, output_dir):<br>    mbox = mailbox.mbox(mbox_path)<br><br>    if not os.path.exists(output_dir):<br>        os.makedirs(output_dir)<br><br>    for i, msg in enumerate(mbox):<br>        subject = get_subject(msg)<br>        body = get_body(msg)<br><br>        file_name = f&quot;{i+1}. {subject[:50].replace(&#39;/&#39;, &#39;&#39;).replace(&#39;:&#39;, &#39;&#39;)}.txt&quot;<br>        file_path = os.path.join(output_dir, file_name)<br><br>        try:<br>            with open(file_path, &#39;w&#39;, encoding=&#39;utf-8&#39;) as txt_file:<br>                email_text = f&quot;Subject: {subject}\n\n{body}\n{&#39;-&#39;*80}\n&quot;<br>                txt_file.write(email_text)<br>                print(f&quot;Success writing file {file_path}&quot;)<br>        except Exception as e:<br>            print(f&quot;Error writing file {file_path}: {e}&quot;)<br><br>mbox_file_path = &#39;_your_.mbox&#39;<br>output_dir_path = &#39;docs&#39;<br><br>export_emails_to_txt(mbox_file_path, output_dir_path)Where _your_.mbox is the name of your saved .mbox file, afterward, I saved the mbox.txtas a box.pdf.</pre><p>While waiting for the Takeout from Google, I also compiled in parallel my most essential presentations, concepts, manuscripts, invoices, and bank statements as PDFs. Finally, I saved them together with the <strong>.mbox</strong> export of around 30.000 mails in a folder called <strong>/docs</strong>.</p><h4>3. API &amp; App</h4><p>To use the OpenAI API, you need a private API key. You can create the key at <a href="http://platform.openai.com/signup">platform.openai.com/signup</a>.</p><p>Now open your editor (like <a href="https://apps.apple.com/us/app/bbedit/id404009241?mt=12">BBEdit</a>) and copy the following code into the editor (or use my <a href="https://github.com/mattesmattes/pca">GitHub repository</a>). You must replace the placeholder <strong>Your API Key</strong> with your Secret API Key, which you obtained from OpenAI.</p><p>Save the script — for example, with the name <strong>pca.py</strong> — in the same place where the <strong>doc</strong> folder with your training data is located.</p><pre>from gpt_index import SimpleDirectoryReader, GPTListIndex, GPTSimpleVectorIndex, LLMPredictor, PromptHelper, Document<br>from langchain.chat_models import ChatOpenAI<br>import gradio as gr<br>import sys<br>import os<br>import codecs<br>from unidecode import unidecode<br><br><br>os.environ[&quot;OPENAI_API_KEY&quot;] = &#39;Your_API_Key&#39;<br><br>def construct_index(directory_path):<br>    max_input_size = 4096<br>    num_outputs = 2048<br>    max_chunk_overlap = 20<br>    chunk_size_limit = 600<br><br>    prompt_helper = PromptHelper(max_input_size, num_outputs, max_chunk_overlap, chunk_size_limit=chunk_size_limit)<br><br>    llm_predictor = LLMPredictor(llm=ChatOpenAI(temperature=0.7, model_name=&quot;text-davinci-003&quot;, max_tokens=num_outputs))<br><br>    documents = []<br>    for filename in os.listdir(directory_path):<br>        with open(os.path.join(directory_path, filename), &quot;rb&quot;) as f:<br>            data = f.read().decode(&quot;utf-8&quot;, &quot;replace&quot;)<br>            # Replace non-ASCII characters with their closest ASCII equivalent<br>            data = unidecode(data)<br>            document = Document(data)<br>            document.metadata = {&quot;filename&quot;: filename}<br>            documents.append(document)<br><br>    index = GPTSimpleVectorIndex(documents, llm_predictor=llm_predictor, prompt_helper=prompt_helper)<br><br>    index.save_to_disk(&#39;index.json&#39;)<br><br>    return index<br><br>def chatbot(input_text):<br>    index = GPTSimpleVectorIndex.load_from_disk(&#39;index.json&#39;)<br>    response = index.query(input_text, response_mode=&quot;compact&quot;)<br>    return response.response<br><br>iface = gr.Interface(fn=chatbot,<br>                     inputs=gr.inputs.Textbox(lines=7, label=&quot;Enter your text&quot;),<br>                     outputs=&quot;text&quot;,<br>                     title=&quot;Your Personal Chat Appliance&quot;)<br><br>index = construct_index(&quot;docs&quot;)<br>iface.launch(share=True)</pre><p>In your terminal program, you move to the directory where your <strong>pca.py</strong> script and the folder with the training data are located and start the script:</p><p><strong>python3 pca.py</strong></p><p>Now the program analyzes the documents using the OpenAI LLM model and indexes the information. In the script, I chose <strong>text-davinci-003</strong> as the model, but you can change it to <strong>chatgpt-3.5-turbo</strong>, too.</p><p>A local vector index of your documents is built and stored in <strong>index.json</strong>. Although we use the specific GPT-API to construct the model, OpenAI promises not to use this data for AI training. You can test the different models by exchange your pre-generated <strong>index.json</strong> while the app is running.</p><p>Processing may take some time, depending on the size of the documents. Start with small amounts to test. Once the LLM has processed the data, you will get some warnings you can ignore. Finally, you will find a local URL at the end. Copy this and open it in a browser.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Z3f_tyIwVsUJLP2F2Qvq4Q.jpeg" /></figure><h4><strong>And what are the results?</strong></h4><p>For coherent documents, the results are promising, as expected. However, this version 0.1 still needs to have session binding, so the conversational benefits known from ChatGPT needs to be added. The outcome from my mailbox archive is very mixed. The model understands individual aspects, but the overall context and the references from thousands of emails are not linked. I need to continue working on this.</p><p>Also, the shift from OpenAI to (Meta’s) <a href="https://til.simonwillison.net/llms/llama-7b-m2">LLaMa models</a> is still a to-do for me: having everything local and saving vast amounts of API costs.</p><p><em>Have fun experimenting!</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f6aeeedc7654" width="1" height="1" alt=""><hr><p><a href="https://medium.com/thymos-x-mimesis/how-i-built-my-very-first-personal-chatbot-appliance-pca-powered-by-chatgpt-and-my-mac-mini-f6aeeedc7654">I built my very first Personal Chatbot powered by OpenAI and my Mac mini, Here is How</a> was originally published in <a href="https://medium.com/thymos-x-mimesis">VRONT</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[ChatGPT-4 may be the best thing that could happen to Apple]]></title>
            <link>https://medium.com/thymos-x-mimesis/chatgpt-4-may-be-the-best-thing-that-could-happen-to-apple-3432fb33ba12?source=rss----ddd852b9eaa8---4</link>
            <guid isPermaLink="false">https://medium.com/p/3432fb33ba12</guid>
            <dc:creator><![CDATA[Matthias "Mattes" Schrader]]></dc:creator>
            <pubDate>Wed, 29 Jan 2025 10:12:03 GMT</pubDate>
            <atom:updated>2023-03-19T08:04:51.380Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="Midjourney prompt: “medium-full shot of Apple’s CEO chatting with an Avatar on an iPhone, Avatar is Tim Cooks Alter Ego designed by Pixar, Apple Park‘s Keynote stage, high resoltion textures and detailed skin texture, 85mm lens with 1:1.2, side-angle view, 4k — ar 16:9 — stylize 1000 — v 5”" src="https://cdn-images-1.medium.com/max/1024/1*N1WXWLywX3RmKDQRq9y5CA.png" /></figure><h3><strong>In the current AI hype, Apple is the only player that seems strangely reticent, while Google, Meta, and Microsoft are beating the drum for their AI-boosted products weekly. And yet, Apple may be the winner of AI’s breakthrough based on Large Language Models (LLMs). Here’s what the Playbook could look like.</strong></h3><p>Apple has redefined the personal computer thrice: with the Apple II, it invented the category in the first place (1977). The user-friendly internet in the gestalt of the WWW was developed on Steve Jobs’s NeXT computers in the early 1990s, and in 2007 with the iPhone, the PC became an everyday companion. Apple evolved into the most valuable company in the world.</p><p>With LLMs, Apple could reinvent the idea of the PC for the fourth time and truly make it a <em>personal </em>computer. But not as a Copilot equipped with generic world knowledge like Microsoft. <strong>Apple has the chance to create a digital alter ego of every human being</strong>. The core here is a <em>personal</em> LLM that is trained by all of its user’s interactions with an Apple device; everything you</p><ul><li>write and read (emails, chats, search, social, texts)</li><li>say and hear (phone &amp; video calls), and</li><li>experience (on the screen, photos, videos, microphone).</li></ul><p>Even if this reads quite spooky in the first (OK, and second) moment, the personal benefit of such a trained omniscient companion would undoubtedly be extremely high. It would know its user extremely well (maybe better than the user knows himself — but that’s another discussion) and would be perfect assistance. The main argument against such a trained LLM is <em>Privacy</em>. Who would want to store their entire life (with all their intimate secrets) in the cloud and leave it to third parties for monetization or leaks?</p><p>But that is precisely Apple’s unfair competitive advantage. Only Apple could integrate large language models into its ecosystem like this without compromising privacy:</p><ol><li>With Apple Silicon (MX processors), Apple brings everything it needs to implement LLMs so that most processing occurs on the personal device. Hence, no training data must be transferred to the cloud (neural engine, GPUs, frameworks).</li><li>Apple would use federated learning techniques to share model updates without exposing raw user data for model training and improvement. By aggregating these updates from many devices, the LLM could be continuously refined while protecting user privacy.</li><li>Unlike the other Big Techs, Apple has put privacy consistency at the center of its communications and doing in recent years. Apple has an almost unassailable trust advantage: it’s <em>the</em> privacy brand today.</li><li>Apple can integrate its AI services not only into existing services (Messages, Mail, Apps) but also across the entire hardware (iPhone, iPad, Mac, AirPods, Watch, AppleTV, Glasses (?)) in a privacy-compliant manner.</li><li>Apple does not need monetization in its business model of personal data like Google, Meta, and meanwhile also Amazon.</li></ol><p>To connect Apple’s ecosystem with this new <em>egosystem</em>, the question remains whether on-device processing of LLMs is possible. Don’t models like GPT-X with billions of parameters require massive cloud infrastructures for this? Yes and no. There are practicable ways to solve this challenge:</p><ol><li><em>Model compression:</em> techniques such as pruning, quantization, and knowledge distillation can be used to reduce the size of LLMs without significant performance degradation. These compressed models are better suited for on-device processing, although they may not be as powerful as their larger counterparts.</li><li><em>FlashAttention:</em> is a new IO-aware exact attention algorithm that speeds up the training of Transformers and reduces its memory footprint — without any approximation. Transformer for actions approaches (like adept.ai) follow this path.</li><li><em>Model partitioning:</em> dividing the LLM into smaller sub-models that work together in a distributed manner can enable parallel processing and reduce the overall computational load on the device.</li><li><em>Prioritization of AI components </em>in Apple’s Silicon Roadmap (neural engine, GPUs, RAM).</li><li><em>Adaptive models:</em> developing models that adapt their complexity to the available resources can help balance performance and resource usage. For example, if a device has limited computing power, the model could switch to a simpler version to ensure smooth operation.</li></ol><p>The recent leak of Meta’s LLaMa model showed <a href="https://arstechnica.com/information-technology/2023/03/you-can-now-run-a-gpt-3-level-ai-model-on-your-laptop-phone-and-raspberry-pi/">that LLMs might perform very well on local devices</a>. Model partitioning or hybrid approaches could make personally trained LLMs in conjunction with generic cloud-based LLMs an unbeatable tandem.</p><p>Apple’s advantage is obvious: the lock-in via personal LLMs on Apple devices would go over the entire lifetime of the user and diffuse into all conceivable hardware. Maybe, Apple will become the first 10-trillion-dollar company after all.</p><p><em>Disclaimer: I assume Apple (like the rest of the world) was surprised by the success of OpenAI’s ChatGPT. And, of course, I have no — zero! — an idea about what is being worked on in Apple Park and the Apple offices scattered all over Cupertino. So everything written down here is my ChatGPT4-like hallucination.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3432fb33ba12" width="1" height="1" alt=""><hr><p><a href="https://medium.com/thymos-x-mimesis/chatgpt-4-may-be-the-best-thing-that-could-happen-to-apple-3432fb33ba12">ChatGPT-4 may be the best thing that could happen to Apple</a> was originally published in <a href="https://medium.com/thymos-x-mimesis">VRONT</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Accenture Song — Why we skip our iconic agency brands]]></title>
            <link>https://medium.com/thymos-x-mimesis/why-we-skip-our-iconic-agency-brands-766b97e36a6c?source=rss----ddd852b9eaa8---4</link>
            <guid isPermaLink="false">https://medium.com/p/766b97e36a6c</guid>
            <dc:creator><![CDATA[Matthias "Mattes" Schrader]]></dc:creator>
            <pubDate>Wed, 29 Jan 2025 10:11:42 GMT</pubDate>
            <atom:updated>2022-05-03T09:15:41.467Z</atom:updated>
            <content:encoded><![CDATA[<iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FKmTVSMnxpK4%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DKmTVSMnxpK4&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FKmTVSMnxpK4%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/6fa81d06dd0aff21c6a96e6f33f60ee9/href">https://medium.com/media/6fa81d06dd0aff21c6a96e6f33f60ee9/href</a></iframe><h3>Accenture Song — Why we skip our iconic agency brands</h3><p><strong>I’ve been repeatedly asked two questions over the last week. Why Accenture Song, and why are we doing this at all? The first question is surprisingly easy to answer, and the second is essential to discuss.</strong></p><p>About the name: I don’t think it’s the best name in the universe, but Song is okay. It’s associative, catchy, upbeat, and fresh. Plus surprising in the context of Accenture. At least our peers are talking around it eagerly and giving us reach. Everyone in the industry knows the name now. Nice hook.</p><p>What I find more stunning is how names are often massively overrated, even by marketing people. While Apple is the most valuable brand name globally, names like Pear or Banana don’t even have the value of a <a href="https://www.apple.com/shop/product/MM6F3AM/A/polishing-cloth">polishing cloth for Apple’s new Displays</a>. It’s not the name but what people associate with a name. To give Accenture Song meaning, we must deliver spectacular products and work. What’s not enough is to talk about how we could make great things. We have to create and ship them. It’s on us — forget the name.</p><p>But why are we sunsetting iconic agency brands like Fjord, Kolle Rebbe, Mackevision, and SinnerSchrader? We’re taking this radical step because our world has radically changed.</p><h3>The explosion of channels is blowing up traditional agency genres.</h3><p>Our clients’ world is changing dramatically. Over the last thirty years, marketing departments have fanned out along the customer touchpoints. For their specific needs, the agency industry has carved out corresponding genres: TV &amp; print Creative, Content, Design, Direct, Event, Media, PR and, of course, Digital. The different agency categories have each offered a clearly defined product: i.e., a campaign or an e-commerce platform. However, the web has compressed these customer touchpoints to a few inches of screen size. The smartphone absorbs all traditional channels as a universal simulation machine and continuously blends up-and-coming media such as games, social, messenger, and metaverse. Our clients’ marketing departmental walls are shrinking accordingly, and every agency category is transforming into a full-circle digital agency today to stay relevant for the CMO. The boundaries between agency genres are becoming blurred. Therefore, it is not without reason that we merged the content teams of Kolle Rebbe and SinnerSchrader this fiscal, for example.</p><h3><strong>Every digital touchpoint becomes a point of sale.</strong></h3><p>Furthermore, clients are crushing the frontiers between marketing and sales. Every digital touchpoint is becoming a point of sale, and businesses transform themselves into direct-to-consumer brands. The new trilogy of branding, performance marketing, and commerce is reorchestrating the corresponding teams in the client organization, and agencies have to rearrange their talent with the same beat. The revisited sales funnel must be designed, created, built, and optimized in a close loop. Any boost in optimizing the experience and achieving relevance in brand, product, and targeting ensures excessive growth in a media world that regulates customer access via an auction mechanic. Only the agency model that masters the close-loop in this marketing/commerce game will grow. The more significant the digital sales channel becomes for our clients, the more we have to say goodbye to the agency classics of yesterday in our industry.</p><h3><strong>Tech rules. Period.</strong></h3><p>Not only traditional agency genres are merging, but agencies also must incorporate technology services. Marketing, commerce, and product innovation are technology plays. Many businesses face the challenge that they have often lost their technology expertise by becoming used to purchasing tech services as a standardized offshore product at the lowest possible price. In front of the customer, nevertheless, standard means commodity. And commodity brands are drowning in the sea of mediocrity today.</p><p>Relevance to the customer can only be achieved through innovation and differentiation — spoken in tech terms, a high degree of custom software is mandatory. Most CIO offices and procurement people have unlearned this. Instead, bespoken software engineering is still too often confused with agile methods. As a result, many companies are trapped in <a href="https://medium.com/tech-sojourna/7-things-wrong-with-deloittes-agile-tube-map-641192e20068">dysfunctional process monsters from hell</a>. In parallel, in-house technology expertise comes through the backdoor via marketing and progressive business units. More and more clients are emancipating themselves from the sweet drug of standard software.</p><h3><strong>At the speed of life.</strong></h3><p>The requirement for a cross-channel view — and therefore infrastructure — of all data plus global roll-outs of our solutions for the clients in our region also gravitate toward a unified brand presence beyond our local labels. But much more important is another driver: speed.</p><p>The pandemic has radically changed customer behavior. The e-commerce volume has grown as much in the last 20 months as it did in the first 20 years of the Internet from 1995 to 2015. We also see tectonic shifts in qualitative terms. The aggregate market share of small and niche brands has exploded from under 20% in the physical world to over 60% in digital channels. Last data point: Shein has unseated Zara and H&amp;M within two years as a pure player from China — but without a business in China.</p><p>The world is changing at a breathtaking pace right now, and David Droga calls it “at the speed of life.” That’s exactly how our clients are changing and why we have to shuffle our playlist.</p><h3><strong>Shuffle and play a new song.</strong></h3><p>We’ve spent the last few months analyzing the clients we serve in Austria, Germany, and Switzerland. There are no two businesses that buy the same things from us. Each client tries to find its answer to the above challenges, for example, which services are built internally as a core competency and which things are cast externally. There are also many discussions about joint ambitions to overcome the traditional procurement processes, which usually boil down to standardization, commoditization, and irrelevance in the customer space.</p><p>Furthermore, our analysis uncovered that we are working together in more and more client engagements across all parts of FKA Interactive. In new business, this share continued to grow strongly last year. At the same time, we discovered that the halo of our strong agency brands implies a limitation in the perception of our capabilities as a group. As I described before, this happens because the agency brands are cut along the buying patterns of the clients’ historical marketing organizations. Yes, the agency brands continue to have their fan base, but that fan base is ageing and shrinking.</p><p>With the fast-forward transformation of the marketing world — expansion of the channels and implosion on the small screen, merging of marketing and sales, the incubation of innovation, delivery of tech platforms and services, and the force of data — the landscape of agency habitats as we know it is diffusing. We have to shuffle our offering for each client across all our capabilities every time. We might have called it custom agencies in the past. Today, I prefer the definition crossover instead: cross capabilities and over the edges of what agencies do (especially in consulting, innovation, and deep tech).</p><p>That’s why we decided to skip our previous agency genres and brands.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=766b97e36a6c" width="1" height="1" alt=""><hr><p><a href="https://medium.com/thymos-x-mimesis/why-we-skip-our-iconic-agency-brands-766b97e36a6c">Accenture Song — Why we skip our iconic agency brands</a> was originally published in <a href="https://medium.com/thymos-x-mimesis">VRONT</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Culture, Context, and Commerce — the Trinity for Brand Growth revisited through the Lens of Web3.0]]></title>
            <link>https://medium.com/thymos-x-mimesis/culture-context-and-commerce-the-trinity-for-brand-growth-revisited-through-the-lens-of-web3-0-f9e0cb88dda8?source=rss----ddd852b9eaa8---4</link>
            <guid isPermaLink="false">https://medium.com/p/f9e0cb88dda8</guid>
            <category><![CDATA[sinnerschrader]]></category>
            <category><![CDATA[accenture-interactive]]></category>
            <category><![CDATA[brand-strategy]]></category>
            <category><![CDATA[web3]]></category>
            <dc:creator><![CDATA[Matthias "Mattes" Schrader]]></dc:creator>
            <pubDate>Wed, 29 Jan 2025 10:10:43 GMT</pubDate>
            <atom:updated>2022-01-04T17:38:47.977Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*QURvhq5twSGpBa-nPDjMiQ.jpeg" /></figure><h3>Culture, Context, and Commerce — the Trinity for Brand Growth revisited in the Age of Web3.0</h3><p><strong>The past year was extraordinary in many ways. Personally, I have learned some new things, reframed already learned, and tried to unlearn some old stuff. Here’s my shot of what’s coming 2022+ and why I’m super optimistic about what lies ahead.</strong></p><p><a href="https://nextconf.eu/2021/01/happy-new-year-2026/#gref">12 months ago</a>, I wrote about how we jumped into a “remote first” world through Covid-19. To illustrate, I borrowed an analogy from the history of the automobile. The first 50 years of the automobile industry — from Carl Benz’s patent motor car in 1886 to the construction of the first Autobahn in 1936 — was about starting car companies and figuring out what cars should even look like and what they should be. The second 50 years, on the other hand, were about designing a world around the everywhere PC (personal car). Only then was it possible to move to the suburbs, work in the city and do your weekend shopping in the middle of nowhere.</p><p>In our industry, the first 50 years began with <a href="https://en.wikipedia.org/wiki/Apple_I">the invention of the PC in 1976</a>. Through Covid-19, we jumped into the second half of the cycle a few years earlier. Accenture’s Julie Sweet calls it <a href="https://fortune.com/videos/watch/How-companies-are-pivoting-their-AI-efforts-in-the-era-of-compressed-transformation-/f4accb9d-749a-43a7-9600-17747c939677">compressed transformation</a>. The roads (cloud) are paved, and the cars (PCs/smartphones) are nearly perfect and ubiquitous. However, unlike the automotive era, the directional vector has turned 180 degrees. People are no longer driving to things, but things are coming as data to them to work, communicate and consume.</p><p>My conclusion culminated in the imperative: everything that can be done remotely will be done primarily remotely in the future.</p><p>How will the world be shaped in the remote first age? How can we help our clients succeed and grow — will it only be about the Metaverse and Web 3.0 in the future? Not quite, and my versioning of the Internet looks a little different.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LOym44z2deH2yKzAwWnmeg.jpeg" /></figure><p><strong>Internet 1.0: Commerce (1994-)</strong></p><p>On Internet 1.0, we create relevance for clients by making them transactional to their customers. We develop commerce platforms, integrate their products into marketplaces, and run appropriate marketing promotions and processes. <strong>We transform clients into direct-to-consumer businesses.</strong></p><p><strong>Internet 2.0: Context (2007-)</strong></p><p>On Internet 2.0, we create relevance by contextualizing brands. We develop communication and content formats for owned and paid channels. Embracing the Internet 1.0, every digital touchpoint becomes an entry point for a transaction. <strong>We transform clients into direct-to-consumer brands.</strong></p><p><strong>Internet 3.0: Culture (2021-)</strong></p><p>On Internet 3.0, we create relevance by seeking and reinforcing cultural resonance for brands. This has been the key insight for me over the past 12 months. In Transformational Products (<a href="https://www.amazon.de/kindle-dbs/entity/author/B06XT2S42Q?_encoding=UTF8&amp;node=186606&amp;offset=0&amp;pageSize=12&amp;searchAlias=stripbooks&amp;sort=author-sidecar-rank&amp;page=1&amp;langFilter=default#formatSelectorHeader">Hardcover/Kindle</a>) I was still justifying behavioral changes of people exclusively through the transformational aspects of digital products (such as 10 x better experiences). I now feel that this approach falls short, and we also need to consider psychological and sociological dimensions. In this sense, we must support our clients not only to be transactional in contextually enriched channels but also to innovate products that are highly fit into today’s culture. Embracing Internet 1.0 and 2.0, every digital product could become a culture-changing artifact. <strong>We transform clients into culture-shaping brands.</strong></p><p>Three terms are essential here: culture, relevance, and resonance. With <a href="https://en.wikipedia.org/wiki/Clifford_Geertz">Clifford Geertz</a> and Max Weber, we understand humans as living entangled in self-spun webs of meaning and these webs we call <strong>culture</strong>. Relevance reflects the meaning of a brand in human minds. With the words of <a href="https://de.wikipedia.org/wiki/Hans_Domizlaff">Hans Domizlaff</a>, <strong>relevance</strong> implies for brands “to secure a monopoly position in people’s psyches.” Following <a href="https://de.wikipedia.org/wiki/Hartmut_Rosa">Hartmut Rosa</a>, <strong>resonance</strong> ultimately characterizes the relationship of people and brands as a vibrating cultural system with significant feedback loops.</p><p>In this logic, innovative products that generate resonance become culture-shaping artifacts: culture cast in code. They must be designed to be as popular as possible to work immediately without intermediate stages or mental effort. The starting point for creating these products is always a deep understanding of people and their specific cultural context. This approach diametrically separates the design and build process from traditional software development, which starts with legacy systems and business processes and ends with users. In contrast, we start the other way around with the experience and context and develop from there back towards technology and processes.</p><p>Digital products — compared to physical products — are only successful if people use them regularly. Sustainability is build-in by design. Digital products create the most value by network effects the more frequently and the more people they are used to. They then resonate with the culture and give it direction. <strong>The transmission belt that connects people and brands is a cultural consensus on shared values.</strong></p><p>If this consensus is missing, something in the system breaks. This can be observed in particular in the hype topics of the past year. Web 3.0 is driven by the idea of decentralization and often involves blockchain technologies, such as cryptocurrencies and non-fungible tokens (NFTs). It’s about building assets (money, rights, code, contracts, etc.) in the form of tokens into the inner workings of almost everything we do digitally. This goes all the way to concepts of decentralized autonomous organizations (DAOs). All this without the involvement of central institutions.</p><p>At the core, cultural battles are being fought here:</p><p>There is a deep distrust of one part of the society against central institutions (from <a href="https://en.wikipedia.org/wiki/Big_Tech">Big Tech</a> to government actions on Covid-19 and vaccinations). This anti-establishment attitude has its roots in both <a href="https://www.wired.com/story/the-libertarian-logic-of-peter-thiel/">the libertarian</a> and <a href="https://music.youtube.com/playlist?list=OLAK5uy_lbSO2Gn3TAK69uG6HP0fTnlLYsohbtF_Y">left-wing cultures</a>.</p><p>In contrast, other parts of society supporting central and vital institutions rely on a democratically legitimized mandate to shape. Here, <a href="https://www.stephendiehl.com/blog/web3-bullshit.html">Web 3.0 is sometimes even judged as</a> “a new anarcho-casino-capitalism world where every fourteen-year-old kid can launch a fly-by-night Ponzi scheme and pump it on social media all from the comfort and anonymity of their parent’s basement.”</p><p>Examples such as the <a href="https://www.theverge.com/22824387/bored-ape-yacht-club-nft-party-new-york">Bored Ape Yacht Club NFTs</a> show that these conflicts revolve around different cultural phenomena, too. In this case, pop culture resonates exceptionally with a digital and financial elite seeking new exclusive markers of distinction.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ctWb_QnzyaSy7mdMceEvQQ.jpeg" /></figure><p>So times are crazy and confusing. But times of cultural upheaval also offer incredible opportunities. From the things just described, we can derive a super simple growth formula for our clients:</p><p>The growth of brands is propelled by</p><ul><li><strong>innovative products </strong>at the intersection of code and culture,</li><li><strong>contextual amplification</strong>, and</li><li><strong>transactional</strong> <strong>ability</strong>.</li></ul><p>Brands that focus on this trinity will have massive growth opportunities in the coming years.</p><p><strong>In this spirit, I wish you curiosity, energy, and joy in 2022, and above all: stay safe!</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f9e0cb88dda8" width="1" height="1" alt=""><hr><p><a href="https://medium.com/thymos-x-mimesis/culture-context-and-commerce-the-trinity-for-brand-growth-revisited-through-the-lens-of-web3-0-f9e0cb88dda8">Culture, Context, and Commerce — the Trinity for Brand Growth revisited through the Lens of Web3.0</a> was originally published in <a href="https://medium.com/thymos-x-mimesis">VRONT</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Kodak Momente]]></title>
            <link>https://medium.com/thymos-x-mimesis/kodak-momente-9d136a8a433c?source=rss----ddd852b9eaa8---4</link>
            <guid isPermaLink="false">https://medium.com/p/9d136a8a433c</guid>
            <category><![CDATA[management]]></category>
            <category><![CDATA[digital-transformation]]></category>
            <dc:creator><![CDATA[Matthias "Mattes" Schrader]]></dc:creator>
            <pubDate>Wed, 29 Jan 2025 10:08:22 GMT</pubDate>
            <atom:updated>2017-05-01T13:38:05.741Z</atom:updated>
            <content:encoded><![CDATA[<h4>Ein Auszug (1/5) aus dem Buch <a href="https://medium.com/still-day-one/transformationale-produkte-1c5e3b4bc791">“Transformationale Produkte”</a></h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*3m2l-_GxlHmz3kWgOFHeeQ.png" /></figure><blockquote>„As one industry after another looks at itself in the mirror and asks about its future in a digital world, that future is driven almost 100 percent by the ability of that company’s product or services to be rendered in digital form.“<br>– Nicholas Negroponte, Being Digital (1995)</blockquote><p>Dass Bits gegenüber Atomen unglaublich viele Vorteile haben, ist keine neue Erkenntnis. Nicholas Negroponte hat dies schon in „Being Digital“ beschrieben. Bereits damals war der grundlegende Zusammenhang erkannt. Es ist regelmäßig um Größenordnungen günstiger, Bits zu verarbeiten und zu distribuieren als Atome. So entsteht eine Digitalisierungsrendite, ohne dass sich an Geschäftsprozessen, Geschäftsmodellen oder Produkten irgendetwas ändern würde.</p><p>Doch dabei bleibt es in der Regel nicht. Das Netz hat die Geschäftsgrundlage drastisch verändert. Die Distributions- und Transaktionskosten digitaler Güter sind praktisch gleich null, was völlig neue Geschäftsmodelle ermöglicht. Die Grenzkosten digitaler Güter sind ebenfalls gleich null, weshalb digitale Güter nicht den Gesetzen der Knappheit unterliegen. Einmal produziert, können sie zu vernachlässigbaren Kosten beliebig oft kopiert werden. Das Internet ist, auch technisch, eine einzige große Kopiermaschine.</p><p>Wer versucht, auf Güterknappheit basierende Geschäftsmodelle zu digitalisieren, muss daher gegen fundamentale Kräfte ankämpfen. Dies kann nur für Produkte funktionieren, die nicht leicht substituierbar sind, und wird ansonsten in den meisten Fällen scheitern. Knapp sind heute Nutzer und ihre Aufmerksamkeit, nicht die digitalen Güter. Deshalb verschieben sich die Machtverhältnisse in der digitalen Wirtschaft zugunsten der Nutzer, Konsumenten oder schlicht: der Menschen.</p><p>In einer ersten Phase Mitte der 1990er-Jahre begann der E-Commerce dank unschlagbar günstiger Distributions- und Transaktionskosten zunächst, den Handel zu revolutionieren. An den Produkten selbst änderte das zunächst nichts. Heute transformiert das Netz hingegen auch die Produkte, indem es zum Kern des Produkterlebnisses wird und häufig Netzwerkeigenschaften den Wert kreieren. Produkte schrumpfen zu einer App für das Smartphone. Die beste <em>User Experience (UX)</em> setzt sich durch, erreicht mehr Nutzer als konkurrierende Angebote und verdrängt diese über kurz oder lang vom Markt. Die User Experience von Uber schlägt nicht nur das bisherige Taxierlebnis, sondern substituiert tendenziell auch den Besitz eines eigenen Autos. Die Atome eines Taxis sind weniger wichtig für das Produkt „Mobilität“ als die Bits, die das Netz von Uber repräsentieren. Digitale Produkte bereichern für den Nutzer die Produkterfahrung, während traditionelle Produkte zur <em>Legacy</em> sowie um Interface und Kundenzugang beraubt werden.</p><p>Durch die Digitalisierung wird sichtbar und greifbar, was neuere Theorien in den Wirtschaftswissenschaften, wie die <em>Service-Dominante Logik</em> <em>(S-DL),</em> seit einigen Jahren postulieren: Für den Konsumenten hängt der Wert nicht am physischen Produkt, sondern am Gebrauchswert, der sich erst durch den Nutzer selbst realisiert. Das eigentliche Produkt ist der Service, der sich digital sehr viel günstiger und schneller darstellen lässt als die Welt der Atome: Digitale Services werden viel schneller viel besser als physische Produkte.</p><p>Ein digitaler Service besteht im Kern aus Software, und Software folgt kürzeren Entwicklungszyklen als Hardware. Amazon aktualisiert seine Software zu Spitzenzeiten mehr als 1.000-mal — pro Stunde. Software-Updates sind aber nicht nur häufiger als neue Hardware, sie werten auch bereits vorhandene Hardware auf. Tesla ist inzwischen dazu übergegangen, die Hardware für autonomes Fahren serienmäßig in seine Autos einzubauen, obwohl die Software dafür noch nicht fertig ist. Sie kann später per Update und gegen Aufpreis aufgespielt werden. <em>Functions-on-Demand</em> sind der stärkste Indikator für diesen Paradigmenwechsel von Atomen zu Bits. Ohne Software ist Hardware wertlos. Zur Beschleunigung trägt zudem die Virtualisierung bei. Dabei wird Hardware durch Software nachgebildet, wodurch — von der eigentlichen Hardware abstrahiert — die Physik effizienter genutzt werden kann. Die Hardware selbst wandert zunehmend in die Cloud.</p><p>Clayton Christensen wies in seinem 1997 erschienenen Bestseller „The Innovator’s Dilemma“ präzise auf den folgenden Widerspruch hin: Für etablierte Unternehmen ist es rational, sich auf ihre profitabelsten Kundensegmente und Produktkategorien zu konzentrieren. Sie ignorieren neue, disruptive Technologien, die nicht die Bedürfnisse ihrer besten Kunden erfüllen oder nicht in ihr bestehendes Geschäftsmodell passen. Neue Technologien sind nämlich oft den etablierten Produkten nur in Teilaspekten überlegen und in anderen Facetten deutlich schlechter. Ihnen gelingt der Markteintritt zunächst oft nur am unteren Ende oder in teuren Nischen.</p><p>Es ist sehr schwer, diese Entwicklung rechtzeitig zu erkennen. In der digitalen Welt entwickeln sich viele Parameter — wie Rechenleistung, Speicher, Bandbreite — exponentiell. Jede Generation verdoppelt ihre Leistung. Da das Ausgangsniveau jeder neuen Technologie bei null liegt, ist der schwerste Schritt der erste — von „Zero to One“, wie es in dem gleichnamigen Buch des PayPal-Mitgründers Peter Thiel heißt. Systeme, die exponentiellen Regeln unterliegen, wachsen zunächst langsam. Erst nach zehn Verdoppelungen ist die Tausendermarke überschritten, doch schon nach weiteren zehn Verdoppelungen die Millionenschwelle und nach noch einmal zehn Verdoppelungen schließlich die Milliardengrenze. Aber das gilt nur im Modell. In der realen Welt wirken auf den exponentiellen Graphen eine dämpfende Sättigungsfunktion und disruptive Abbruchkanten in Form von Wettbewerb. Nur aus dem technologischen Fortschritt allein lassen sich kaum robuste Unternehmen begründen. Andy Grove, der Intel zum erfolgreichsten Chip-Produzenten formte, betitelte seinen Managementbestseller daher zu Recht mit „Only the Paranoid Survive“.</p><p>Solange ein Produkt und ein Geschäftsmodell funktionieren, ist der Veränderungsdruck begrenzt. Gerade große Unternehmen sind extrem gut darin, einmal etablierte Produkte und Geschäftsmodelle sehr systematisch zu verfeinern und inkrementell zu verbessern. Kodak ist dafür ein Paradebeispiel. Das Unternehmen hatte immer sehr hohe Forschungs- und Entwicklungsaktivitäten sowie große Labors und investierte viel in Innovation. Trotzdem wurde es schließlich überrollt von der Digitalisierung, die den chemischen Filmprozess obsolet machte. Das alte Geschäft lief einfach zu lange zu gut. Auch weil Kodak es von Jahr zu Jahr immer wieder schaffte, das bestehende Produkt zu verbessern.</p><p>Hinter der Ablehnung der Digitalkameras steckte nicht zwingend Arroganz, denn weder das Unternehmen noch die Menschen, die dort arbeiteten, waren dumm. (Im Gegenteil: Es sind in der Regel schlaue Menschen, die sehr genau schauen, was funktioniert und was nicht.) Wenn sie ihre Kunden befragten, dann bekamen sie beispielsweise zur Antwort: Wir möchten einen Film haben, der noch leuchtendere Farben hat, sich noch schneller entwickeln lässt und noch unempfindlicher ist gegen Belichtungsschwankungen, wie man sie bei der Fotografie häufig hat.</p><p>Genau diese Produkteigenschaften hat Kodak von Jahr zu Jahr stetig versucht zu verbessern. Schließlich konnte sich kaum ein Konsument vorstellen, wie Fotografie komplett anders funktionieren könnte — nämlich mit einem digitalen Chip. Und verzwickterweise war das digitale Bild am Anfang noch dramatisch schlechter als das chemisch prozessierte Foto, dessen Verfahren über Jahrzehnte hinweg optimiert wurde. Das alte Produkt ist an seinem Höhepunkt fast immer der disruptiven Konkurrenz überlegen, die sich von den Rändern des Marktes heranpirscht.</p><p>Die ersten kommerziellen Digitalkameras lösten viel gröber auf als Spiegelreflexkameras in Kombination mit traditionellen Filmen. Elektroautos sind Verbrennern in Aspekten wie Reichweite und Preis unterlegen. YouTube-Videos waren am Anfang viel schlechter als das Fernsehbild — das Bild ruckelte, der Bildschirm war klein und die Bildqualität schlecht. Trotzdem setzen sich die Streamingformate gegen das lineare TV zunehmend durch. Digitale Produkte haben entscheidende Vorteile: Die grundlegende Infrastruktur entwickelt sich exponentiell, wie wir im Folgenden sehen werden. Durch die Vernetzung und die damit einhergehenden Netzwerkeffekte kommt noch eine zusätzliche Nutzendimension hinzu.</p><p>Der Intel-Mitgründer Gordon Moore prognostizierte in den 1960er-Jahren, dass sich die Transistorendichte von integrierten Schaltungen jährlich verdoppeln würde. Bis heute hat sich diese Prognose erstaunlich lange bewährt, auch wenn sich die Verdoppelung bereits seit den 1970er-Jahren eher bei rund 18 Monaten einpendelt. In dieser Zeitspanne steigert sich die Leistungsfähigkeit von Chips oder Speichermedien um den Faktor zwei — zum gleichen Preis.</p><p>Das exponentielle Anwachsen der Rechenleistung bedeutet gleichzeitig auch den dramatischen Preisverfall von Rechenleistung, Speicherkapazität, Sensorik und Netzwerkbandbreite. Unser Zeitalter der Digitalisierung ist im Kern eigentlich ein Zeitalter der Vernetzung. Durch den Preisverfall diffundierte die Netzwerkfähigkeit von zentralen Groß- und Abteilungsrechnern über Personal Computer zu Smartphones und wird schließlich unter dem Begriff <em>Internet of Things</em> <em>(IoT)</em> ubiquitär. Schon für wenige Eurocent-Beträge kann heute alles ins Netz eingebunden werden. Der Mix aus immer leistungsfähigeren, vernetzten Devices und globalen Cloud-Infrastrukturen befeuert eine Explosion von neuen Services und Produkten.</p><p>Schon vor der Digitalisierung verzweifelten Unternehmen daran, ihr profitables (Alt-)Geschäft durch weniger profitables Neugeschäft und zunächst qualitativ schlechtere Produkte zu kannibalisieren. Seitdem gilt es als gesichertes Wissen, dass disruptive Innovation quasi von außen kommen muss, weil Unternehmen systematisch blind dafür sind und es vielleicht sogar sein müssen. Doch stimmt das eigentlich? Ist es wirklich so, dass Unternehmen im Grunde nur abwarten können, bis jemand von außen kommt und ihr Geschäft ruiniert? Oder können sie disruptive Innovationen auch selbst schaffen?</p><p>— <br>Mehr im Buch, das ab sofort erhältlich ist. Beispielsweise bei <a href="https://www.amazon.de/dp/3981871103/">Amazon</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9d136a8a433c" width="1" height="1" alt=""><hr><p><a href="https://medium.com/thymos-x-mimesis/kodak-momente-9d136a8a433c">Kodak Momente</a> was originally published in <a href="https://medium.com/thymos-x-mimesis">VRONT</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Bring Your Own Team]]></title>
            <link>https://medium.com/thymos-x-mimesis/byot-173dd7ef4d62?source=rss----ddd852b9eaa8---4</link>
            <guid isPermaLink="false">https://medium.com/p/173dd7ef4d62</guid>
            <category><![CDATA[team]]></category>
            <category><![CDATA[sinnerschrader]]></category>
            <category><![CDATA[teamwork]]></category>
            <dc:creator><![CDATA[Matthias "Mattes" Schrader]]></dc:creator>
            <pubDate>Wed, 29 Jan 2025 10:07:46 GMT</pubDate>
            <atom:updated>2020-12-09T19:14:10.066Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*1DnsXDb0AQkCquS4GDtOqA.jpeg" /></figure><p><strong>Wann machen uns Projekte wirklich Spaß? Wann schaffen wir wirklich außergewöhnliche Dinge? Wenn wir darüber nachdenken, lautet die Antwort oft: wenn wir in ebenso außergewöhnlichen Teams arbeiten. Wenn wir uns vertrauen, ergänzen und inspirieren. Warum finden wir das so selten?</strong></p><p>Es ist doch so: wir alle wollen in einem festen und funktionierenden Team arbeiten. In der wirklichen Welt, wird das aber selten unterstützt. Zum Beispiel bei Einstellungen. Obwohl wir wissen, dass das Geheimnis erfolgreicher Projekte und Produkte immer Teams sind, finden sich in unseren <a href="https://sinnerschrader.com/de/karriere/?">Jobanzeigen</a> nur einzelne Stellen. Vielleicht ist das nicht sinnvoll: denn wir arbeiten ja nicht alleine. Daher legen wir bei <a href="https://sinnerschrader.com/de/">SinnerSchrader</a> viel Aufmerksamkeit auf die Zusammenstellung von guten Teams. Und wir wollen sogar einen Schritt weiter gehen und was wirklich Neues ausprobieren (nun gut, es ist eigentlich nur <a href="https://stripe.com/blog/bring-your-own-team">relativ neu</a>). Unter <a href="http://s2-jobs@accenture.com/">s2-jobs@accenture.com</a> könnt ihr Euch jetzt auch als komplettes Team bewerben. Und so funktioniert das:</p><ol><li>Sendet uns an <a href="http://s2-jobs@accenture.com/">s2-jobs@accenture.com </a>Eure LinkedIn/XING-Profile und wo und wann ihr erfolgreich zusammengearbeitet habt.</li><li>Wir suchen Teams von 2–7 Köpfen mit Schwerpunkt in der Softwareentwicklung aber auch interdisziplinäre Teams aus UX, Design und Beratung, die gerne zusammenarbeiten.</li><li>Wir stellen sicher, dass der gesamte Einstellungsprozess (Interviews, Feedbacks, Angebote) immer synchronisiert ist. Wir unterbreiten jedem von Euch im Team ein Angebot — oder keinem.</li><li>Wir stellen sicher, dass ihr bei SinnerSchrader — solange ihr es wünscht — immer zusammen arbeitet.</li><li>Wir haben tolle Büros in <a href="https://www.google.com/maps/place/SinnerSchrader/@53.557094,9.926201,17z/data=!3m1!4b1!4m2!3m1!1s0x47b18591e7504141:0x1a57cf875d28e604">Hamburg</a>, <a href="https://www.google.de/maps/place/Friedenstra%C3%9Fe+91,+10249+Berlin/@52.522018,13.43695,17z/data=!3m1!4b1!4m2!3m1!1s0x47a84e14b942b331:0xd2755cdc766ab217">Berlin</a>, <a href="https://www.google.com/maps/place/SinnerSchrader/@50.1137125,8.7170035,17z/data=!4m6!1m3!3m2!1s0x47bd0bfa1014b789:0xe4bed44fc61a4e8f!2sSinnerSchrader!3m1!1s0x47bd0bfa1014b789:0xe4bed44fc61a4e8f">Frankfurt</a> und <a href="https://www.google.com/maps/place/SinnerSchrader/@48.1520709,11.581572,17z/data=!3m1!4b1!4m2!3m1!1s0x479e744a64524ba5:0xdae97b91020bb279">München</a>. Wenn ihr an einem anderen Ort arbeiten wollt, versuchen wir das zu ermöglichen.</li></ol><p>Wir freuen uns, von Euch zu hören.<br><a href="http://s2-jobs@accenture.com/">s2-jobs@accenture.com</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=173dd7ef4d62" width="1" height="1" alt=""><hr><p><a href="https://medium.com/thymos-x-mimesis/byot-173dd7ef4d62">Bring Your Own Team</a> was originally published in <a href="https://medium.com/thymos-x-mimesis">VRONT</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Ach, Maurice!]]></title>
            <link>https://medium.com/thymos-x-mimesis/maurice-h%C3%B6r-doch-auf-zu-heulen-7fcaef4dce65?source=rss----ddd852b9eaa8---4</link>
            <guid isPermaLink="false">https://medium.com/p/7fcaef4dce65</guid>
            <category><![CDATA[publicis]]></category>
            <category><![CDATA[agenturpreisliste]]></category>
            <category><![CDATA[sinnerschrader]]></category>
            <dc:creator><![CDATA[Matthias "Mattes" Schrader]]></dc:creator>
            <pubDate>Wed, 29 Jan 2025 10:07:10 GMT</pubDate>
            <atom:updated>2016-06-16T16:36:07.618Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9fUO8o9YGp1NK-7Nu3HZfA.png" /></figure><h4>Früher war alles besser. Das Wetter, die Werbung, und die Wertschätzung der Kunden sowieso. Im Februar diesen Jahres hat Maurice Lévy, Grandseigneur der Werbung und als CEO von Publicis Chef der drittgrößten Werbeholding der Welt, seinem Frust über die heutigen Verhältnisse gegenüber dem britischen <a href="http://www.campaignlive.co.uk/article/levy-pay-us-due/1383689">Fachdienst CAMPAIGN</a> freien Lauf gelassen. Es war ein seltener und zugleich seltsamer Einblick in die Seele eines Mad Men.</h4><p>So forderte Maurice Lévy vor allen Dingen eine radikale Abkehr von der aktuellen Abrechnungspraxis für Agenturleistungen. Dieses Modell würdige nicht annähernd den Wertbeitrag von Agenturen für Marken. Agenturen müssten der Falle des Stundenbetriebs entfliehen und wieder Beziehungspartner auf Augenhöhe werden. Schließlich verdienten die Agenturen entschieden zu wenig, die Qualität sei gefährdet und am Ende würden daran die Kunden selbst am meisten leiden. Es sei Zeit für ein neues Modell. Leidenschaftlich plädiere er für eine Basis-Fee mit zusätzlichen Lizenzgebühren, die den Wert des kreativen Outputs widerspiegle.</p><p>In welchem Paralleluniversum lebt eigentlich der gute Mann?</p><p>Zunächst das Offensichtliche: Publicis verdient prächtig. Unter dem Strich liefert die Holding stabil eine zweistellige Rendite auf einem Umsatzvolumen von knapp 10 Milliarden Euro. Von diesen Kennziffern können die meisten Kunden von Publicis nur träumen. Mit welcher Chuzpe fordert ein Dienstleister, deutlich profitabler als die Mehrzahl seiner Kunden sein zu wollen? Nebenbei: Lévy selbst bezog in den letzten fünf Jahren eine Gesamtvergütung von über 30 Millionen Euro — was ihn nicht unbedingt zur Idealbesetzung für eine Round-Table-Runde über prekäre Agenturvergütungen macht. Zudem gehört Publicis zu den aggressivsten Akquisiteuren am Markt: Allein im letzten Jahr addierten die Franzosen rund 3 Milliarden Euro Umsatz in ihre Bücher. Knappe Kasse fühlt sich anders an.</p><p>Doch davon abgesehen ist das wirklich Verstörende an Lévis Frustattacke die empfohlene Therapie: Lizenzenerlöse, mit denen Agenturen künftig neben einem Grundrauschen ihre Bilanzen fetten sollen. So ein Modell hat theoretisch überhaupt nur im Bereich der Kreativagenturen Sinn. Aber praktisch? Wer bewertet wie die Lizenzhöhen in einer Welt explodierender digitaler Touchpoints und komplexer User Journeys? Welchen langfristigen Wert haben überhaupt kurzzeitige Inszenierungen und kreative <em>one-hit wonder</em> in einer Zeit, in der Marken zunehmend systemisch agieren, um eine nutzenstiftene Relevanz im Kundenalltag zu bekommen? Was wären schließlich die maßgeblichen KPIs: virale Reichweite, Media Spendings, Awareness-Aufbau, NPS oder gar der neue, von Google und dem ADC definierte Data Creativity Score (DCS)?</p><p>Dabei sind Kreativagenturen bei den großen Netzwerken und so auch bei Publicis oft nur noch ein Nachhall der guten alten Zeit. Mehr als die Hälfte der Umsätze und der Löwenanteil des Wachstums stammen mittlerweile von digitalen Baukonzernen wie Sapient und Razorfish mit Tausenden von Entwicklern, die weltweit für Marken Plattformen und Marketinginfrastrukturen aus dem Boden stampfen. Deren DNA hat mehr mit IBM gemein als mit der von Saatchi &amp; Saatchi. Abrechnungsgrundlage sind Stunden oder werkvertragsähnliche Festpreise, die allerdings aufgrund der agilen Moden zugunsten der Stundenmessung rückläufig sind. Lizenzmodelle sind für diese Gewerke kulturelle No-Go-Areas.</p><p>In der verbleibenden Hälfte des Umsatzkuchens von Publicis spielen Mediaagenturen die dominierende Rolle. Auch hier greift die Lizenzidee vollkommen ins Leere. Schlimmer: Sie ist ein Indiz dafür, dass der Wunsch nach einer Value-basierten Bezahlung offenbar weiterhin ausgeprägt ist. Dabei verkennt Lévy, dass die ganzen Performance-getriebenen Modelle in der Media seit Jahren nur noch massive Schäden anrichten, weil sie naturgemäß Intransparenz voraussetzen und damit Misstrauen schaffen. Systemseitig verhindern sie saubere Prozesse, Datenflüsse und konsistente Wirkungsanalysen der Maßnahmen. Kulturell untergraben performance-getriebene Mediamodelle gerade das, was Lévy mit Leidenschaft fordert: Partnerschaft und Vertrauen zwischen Werbetreibenen und Agenturen auf Augenhöhe.</p><p>Die Wahrheit ist: die Abrechnung nach Stunden ist für Agenturen das Beste, was ihnen passieren konnte. Es ist das Modell der Anwälte, Wirtschaftsprüfer, IT-Dienstleister und Unternehmensberater. Es ist einfach, gelernt und adäquat für eine Dienstleistergattung am Ende ihrer Adoleszenz. Zeit, erwachsen zu werden.</p><p>Warum viele Agenturen mit dem Stundenmodell hadern, liegt nicht am Modell, sondern an der Höhe der Sätze. Dies ist das eigentliche Problem, welches regelmäßig durch Honorardebatten überdeckt wird.</p><p>Dass die Branche hier ein Problem hat, lässt sich an einer ganz einfachen Kennziffer ablesen: der Pro-Kopf-Umsatz von Kommunikations- und Digitalagenturen blieb in den vergangenen Jahren stabil, während im gleichen Zeitraum die Gehälter massiv gestiegen sind. Im Digitalbereich übersteigen bereits heute nicht selten die Freelancer-Sätze für einzelne Gewerke, die für selbige verhandelten Rate Cards der Agenturen mit ihren Kunden.</p><p>Wer sich also um die Zukunftsfähigkeit von Agenturen und ihre Fähigkeit sorgt, langfristig Talente zu gewinnen und zu binden, muss daran arbeiten, die Stundensätze zu erhöhen.</p><p>Hierfür stelle ich zwei Maßnahmen zur Diskussion, die von keiner Agentur alleine, sondern nur gemeinsam umgesetzt werden können.</p><p><strong>Zunächst: „Rate first, Pitch second“. </strong>Erst werden die kaufmännischen Rahmenbedingungen geregelt, dann wird gepitcht. Agenturen können nur dann auf Augenhöhe auskömmliche Rate Cards verhandeln, wenn das heutige Vorleistungsprinzip entfällt. Denn hat die Agentur erst einen sechsstelligen Betrag in einen Pitch investiert, hat sie spieltheoretisch kaum noch eine Chance, im nachgelagerten Einkaufsprozess ihre Stundensätze auf einem angemessenen Niveau zu verteidigen.</p><p>Aus Verlustangst, bereits geleistete Pitchaufwände abschreiben zu müssen, lassen sich Agenturen regelmäßig in kundenseitig professionell geführten Einkaufsprozessen auf das Preisniveau leistungsschwächerer Wettbewerber herunter handeln. Agenturen sollten also erst dann in einen Pitch investieren, wenn die kommerziellen Rahmenbedingungen einer späteren Zusammenarbeit vorher geklärt wurden. In der Rückkopplungsschleife bereinigt sich das Teilnehmerfeld, wenn die Konditionen unattraktiv sind.</p><p>Stimmen aber die Bedingungen, sind die Agentur noch stärker motiviert, Vollgas zu geben. Das ist wichtig, denn Pitches sind ein notwendiger Innovationsmotor in unserer Industrie. Jede heute am Markt erfolgreiche Agentur begann als Start-up. Die unsichtbare Kette vom ältesten bis zum jüngsten Schreibtisch besteht aus einzelnen Pitchgewinnen und der steten Beweisführung einer Bestleistung im Wettbewerb. Es sind nicht Goldene Löwen und Effies, die Agenturen prosperieren lassen, sondern harte Pitcherfolge. Nur diese sorgen für die ständige Häutung von Agenturen, die sich so permanent neuen Marktanforderungen anpassen müssen.</p><p>Wettbewerb, Innovation und stete Leistungsbeweise sind die konstitutionellen Eckpfeiler der Agentur-Kunden-Beziehung und wirken kulturell bis in jede Faser einer Agentur: den Kunden im steten Ringen für die richtige Lösung zu begeistern. Pitches sind die magischen Momente, in denen Agenturen nach intensiver Durchdringung von Markt, Marke und Medium nicht nur eine fundierte Antwort aufs Briefing liefern, sondern eine bessere Zukunft für ihre Kunden (und sich selbst) erfinden. Dies leisten keine Pitch-Placebos wie Kennenlern-Workshops oder Testprojekte mit ihrem Bias der zwischenmenschlichen Chemie und zu geringem Tiefgang in der Aufgabenlösung.</p><p><strong>Die zweite Maßnahme ist so einfach wie selten: „Index Rate Cards“. </strong>Die meisten Agenturen arbeiten heute mit Rate Cards, die nicht mit den stets steigenden Arbeitskosten in unserer Branche synchronisiert sind. Seit Jahren stabile Stundensätze sind eher die Regel als die Ausnahme in den meisten Kundenbeziehungen. Eine entsprechende vertragliche Dynamisierung der Sätze auf die steigenden Kosten hätte bereits nach nur wenigen Jahren eine massive Wirkung, würde das gesamte Niveau der Branche heben und Frustattacken auf die Momente beschränken, in denen sie wirklich mal gerechtfertigt sind — bei einem verlorenen Pitch.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7fcaef4dce65" width="1" height="1" alt=""><hr><p><a href="https://medium.com/thymos-x-mimesis/maurice-h%C3%B6r-doch-auf-zu-heulen-7fcaef4dce65">Ach, Maurice!</a> was originally published in <a href="https://medium.com/thymos-x-mimesis">VRONT</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>