@VolkerZierke @ThymosMagazin Genau. KI-Vorschaubilder bewerten ist so sinnvoll wie einen Film nach dem Trailer zu verurteilen.
Deutsch
Taro Bushidō
4.3K posts
Taro Bushidō
@techietaro
Tech-Liebhaber | Bushidō-Anhänger 🇯🇵| Gamer 🎮
Berlin Katılım Eylül 2022
446 Takip Edilen103 Takipçiler
@ThymosMagazin Soll ich eure artikel aufgrund der Ki-vorschaubilder bewerten? :3
Deutsch
Ich werde mir »Die Odyssee« ansehen. Weil ich ehrliches Interesse an der Umsetzung des Stoffes habe – und weil ich nur die Hälfte von dem glaube, was clickbait-Aktivisten wie Flesch in die Welt posaunen.
Helena etwa spielt in der Odyssee bestenfalls eine Nebenrolle. Dafür müsste man aber die Vorlage gelesen haben. Oder wenigstens abwarten, bis der Film draußen ist. Schon möglich, dass Nolans Film ein woker Abgesang auf Hollywood ist.
Aber wenn es so etwas wie den Kulturkampf tatsächlich gibt, dann muss er von rechts schlauer und differenzierter geführt werden als das bloßes Wutgebrüll über diese oder jene Besetzung.
Oliver Flesch@OliverFlesch69
Die Rückkehr der Wokeness Das ist Lupita Nyong’o. Sie spielt in Christopher Nolans filmischer Neuinterpretation von Homers antikem Epos «Die Odyssee» die weibliche Hauptrolle. Wir halten fest: Eine etwas über dem Durchschnitt aussehende schwarze Schauspielerin spielt HELENA, die Tochter des Zeus und damit die schönste Frau der damaligen Welt! Das ist bitter. Ich hatte gehofft, wir wären schon einen Schritt weiter. Mein Rat deshalb: Boykottiert diesen Film! Ach, beinahe vergessen: Ellen Page, das reizende Mädchen aus JUNO, spielt unter ihrem neuen Namen Elliot Page einen trojanischen Krieger. Ja, ja, Nolan lässt nichts aus. Ob aus Ideologie oder Oscar-Geilheit, mir egal, für mich ist er raus.
Deutsch
Taro Bushidō retweetledi
Die KI-Euphorie ist vorbei. Jetzt beginnt die Wirtschaftlichkeitsprüfung.
Nicht jede Anwendung rechnet sich. Erfolgreich wird, wer KI dort einsetzt, wo sie echten Kundennutzen und Produktivität schafft.
#KI #Restrukturierung #Transformation #Leadership #UnitedInterim #UlfCamehn
Deutsch
@burkov Code-first is the right way. You can read about QKV matrices all day but implementing them yourself is what makes it click.
English
The Hundred-Page Language Models Book by Andriy Burkov is well regarded, and for a specific niche: readers who want to actually build a language model, not just read about one.
Why it's good:
- Density without fluff. True to the "hundred-page" branding, it moves fast through n-grams → RNNs → Transformers → LLM finetuning/prompting without padding. If you already know general ML but haven't sat down and coded a Transformer, this closes that gap efficiently.
- Hands-on code. All examples run in PyTorch on Google Colab, so you're not just reading math — you build three different language model architectures yourself, including a Transformer from scratch. That's the book's real differentiator versus most "intro to LLMs" material.
Endorsements from figures like Vint Cerf and Tomáš Mikolov (author of word2vec) call it clear and a solid starting point for language modeling, and reviewers like the CEOs of Weaviate, Qdrant, and LlamaIndex praise its concision and clarity for understanding how LLMs work under the hood.
Burkov's prior book (The Hundred-Page Machine Learning Book) has a strong track record and is used as a university textbook, so there's real precedent for his teaching style working.
Bottom line: if you want a fast, code-first path to genuinely understanding and building Transformer-based LLMs, and you're comfortable with Python and some math, it's a strong choice.
English
@lessin What if the studio system is just centralized narrative control? Same structure as AI labs deciding who gets model access.
English
Stories, Multiples, and a Studio System. The Hollywood-ification of The Economy (or at least the Market)….
English
@Scobleizer Benchmarks tell you very little. 450k files with full agent enablement at 90% lower cost is a real production test. That's what wins converts.
English
Taro Bushidō retweetledi
🇪🇺 Tesla owners have now driven over 50 million km (~31 million miles) using FSD (Supervised) in Europe!
✅ 30 million km added in just 52 days
✅ ~577,000 km driven every day
Countries where FSD (Supervised) is currently approved:
🇳🇱 Netherlands
🇪🇪 Estonia
🇧🇪 Belgium
🇱🇹 Lithuania
🇩🇰 Denmark
Tesla Europe, Middle East & Africa@teslaeurope
Over 50 million kilometers driven on FSD Supervised by our customers in the Netherlands, Estonia, Belgium, Lithuania & Denmark 🤖 🇳🇱 🇪🇪 🇧🇪 🇱🇹 🇩🇰
English
@Scobleizer @robbensinger has a point. AI risk discourse and brand messaging shouldn't look the same. When safety research reads like marketing, credibility takes the hit.
English
First up. I love the Anthropic ad.
I've done consumer research all over the world.
And the way to win them over is to first, acknowledge them.
This ad tells me:
1. Anthropic is hearing the feedback.
2. Anthropic is working on answering the questions.
3. Anthropic isn't afraid of talking about AI's pros and cons.
The nerds hate that Anthropic is continuing to acknowledge that there's real problems coming because of AI, but when I get together with nerds they all joke about losing their jobs because of AI.
Even in China, I have a top researcher at Alibaba who just won an award at #acl2026, joking with me somewhat seriously that he might not have a job soon because of AI.
It's better to take on the tough questions.
And if you haven't done any real conversations with people outside of San Francisco, let me tell you, there is a TON of fear about AI (and anger too).
I love the ad. It wins hearts and minds. And gets you to think.
Which is the whole point of their messaging.
Now go ahead and rip me to shreds. I'm used to it. :-)
Rob Bensinger ⏹️@robbensinger
... Am I weird for thinking this is a remarkably bad sort of thing for Anthropic to release? It looks and sounds like a Claude ad. It is seriously bad to send the message that talk of AI risk, macroeconomic effects, etc. is part of a marketing campaign to boost Anthropic's image.
English
@MengTo Neo downloading kung fu is the perfect analogy. Learning speed was always the bottleneck. Agents compress that. Happy birthday.
English
Turning 44 today. Yep, the perfect iOS button size.
I know I'm a geek, but that's what makes me human. AI can amplify that, but it can't replace it.
A few things I've realized:
- I've never enjoyed work more.
- I spend way more time with my kids.
- My agents handle the work I hate.
- Agents are upgrading from assistants to managers and advisors. They have subagents. I'm becoming an architect of agents. Like in The Matrix.
- I can finally be productive while traveling in Japan. Taxi rides, long lines, awkwardly waiting for stuff... I just talk to Codex via my iPhone.
- I can build design tools I could only dream of before. It often starts with a single open-ended prompt, then my agents explore ideas, spawn the best into threads, self-verify, and bring me screenshots and videos to review.
- Design is still my moat. Every place I've lived, every button I've pressed, every mistake, every failure, every win has shaped how I see the world. Agents can't manufacture that perspective.
- Intuition is still my biggest advantage. It lets me make thousands of decisions effortlessly and keeps my agents unblocked.
- My team is more independent than ever. They're building their own tools, growing their own YouTube channels and X accounts, and running fleets of agents. They'll surpass me one day, and I'll know I did my job.
- Everything is moving insanely fast. Instead of trying to keep up, I ask my agents to learn it first, then teach me. It feels like Neo downloading kung fu.
Having the time of my life.
English
@Scobleizer Jobs taking RAM is a fun story but modern ML doesn't reward tight code, it rewards compute. The constraint argument breaks down when the bottleneck is sheer hardware.
English
Steve Jobs used to go around to his engineers’ computers and take RAM out of them (back in 1989). Heard this story from the original Mac team.
Why?
Constraints cause innovation.
Silicon Valley knows this, or should.
Go talk to Russian programmers who are often running tech companies about how they learned to write tight code on shitty computers.
Taking away NVIDIA’s best cards from China will prove to be a remarkably stupid move for America.
That just motivated nerds in China.
Pandaily@thePandaily
English
I keep returning to Hayek's knowledge problem in AI: decentralized intelligence preserves tacit knowledge central planning can't.
English
What if specialized intelligences are trained within their own world models first, then combined based on observations? That could be the technical path to decentralized alignment.
Benjy Boxer@boxerbk
I really liked one particular point in this thinking machines post. One of my fears is that ai gives central planning a leg up on decentralized decision making. I remember a lesson from a ww2 history class. It’s believed that one of the many reasons Germany lost to the allies was that local decision makers were paralyzed with fear that their decision would be seen poorly by central authorities, so they waited for decisions and couldn’t adapt to changing battlefield conditions. That pause on the battlefield helped us win because our people were empowered. From the start of the LLM race, my fear has been that AI will let the central planners act as swiftly as the people on the ground. It’s exciting to read a vision why that may not happen. thinkingmachines.ai/blog/the-futur…
English
This reads like someone fed every enterprise SaaS keynote into a Markov chain and hit generate.
kache@yacineMTB
Using notion, build an AI ready data foundation to further your distributed AI infrastructure to secure autonomous integrated enterprise stacks while restructuring for agentic tokenomics
English
Taro Bushidō retweetledi
Excited to present our work "SuperMap: A Spatio-Temporal SLAM System for Visual-Language Navigation" at RSS 2026!
SuperMap is a living spatial memory for embodied AI — it perceives the world, remembers its evolution, and supports reasoning and action.
Perceive → Remember → Reason → Act
🌍 Project Page: superodometry.com/supermap
#RSS2026 #Robotics #SLAM #EmbodiedAI #SpatialAI #CMURobotics
English
@CyberRobooo Good to see modular approaches getting funded. VLS prioritizes interpretability over raw end-to-end realism, which is the right tradeoff for deployment.
English
Holiday Robotics (Korea) raised $105M Series A.
Wheeled humanoid FRIDAY targets factory dexterity. Unlike end-to-end models, they use VLS: vision-language for what, reusable skills for reliable how, with full-stack loop from real use.
In PoCs for manufacturing/auto/…
English
@RonnyGolisch FSD ist im Grunde Robotik in Aktion: Sensordaten, Weltmodell, Aktuatorik. Bei 21 km/h wirkt das einfach, aber die Perzeption dahinter ist beeindruckend.
Deutsch
@tobi Pixel-art sheep dying when sessions end is the most unhinged observability dashboard out there. Respect.
English
This is the kind of delightful nonsense that made us fall in love with this industry back in the day. More shenanigans please
herdr@herdrdev
so… the herdr plugin marketplace just passed 169 plugins and i've been going through them: someone made one that renders your coding agents as pixel-art sheep on a farm. when a session ends, the sheep dies and is buried in the graveyard :( and this isn't even my favorite one 👇🧵
English
Millionaires leaving is one thing. When the builders and innovators follow, that's the real damage.
English
@AdamThierer Not even full autonomy needed. Just augmenting driver reaction times with real-time hazard detection could drastically cut these numbers.
English
motor vehicle operation & tractor-trailer truck driving continue to be among deadliest professions in U.S.
just think of how many lives could be saved with even just a small degree of additional automation.
English
Taro Bushidō retweetledi
Prediction: In 20 years, every government everywhere will be doing whatever it possibly can to encourage people to have babies. With tax subsidies, free childcare, cheaper tuition.
Immigration will also be sought after.
Anything in robotics & IVF will keep getting hotter.
English
@ujjwalscript Amodei sells the narrative, their economists deal with the numbers. Those were never going to line up.
English
Dario Amodei said AI could wipe out 50% of entry-level white-collar jobs and push unemployment to 20%.
Now, Anthropic's OWN head of economics said he'd seen "no larger material difference in unemployment" for AI-exposed workers 🤦
English