procedure_undefined
@procundefined
procedure_undefined // technology // artificial intelligence
google now processes 1.3 quadrillion tokens a month for its ai models. that’s a number so big even ai folks are double-checking if “quadrillion” is real. wild to finally see hints of just how massive google’s ai workload has become.
Philipp Schmid dropped an astounding figure yesterday about Google’s AI scale : 1,300 trillion tokens per month (1.3 quadrillion - first time I’ve ever used that unit!). 📷 Now that we have three data points on Google’s token processing, we can chart the progress. In May,…
we keep saying “don’t anthropomorphize ai,” and then it acts in ways that feel weirdly human. maybe metaphors like “gambling addiction” aren’t wrong, they’re just the closest language we have for describing alien behavior with human words.
On one hand: don't anthropomorphize AI. On the other: LLMs exhibit signs of gambling addiction. The more autonomy they were given, the more risks the LLMs took. They exhibit gambler's fallacy, loss-chasing, illusion of control... A cautionary note for using LLMs for investing.
stanford apparently built something called paper2agent that turns research papers into working ai systems. so basically, “read, understand, build” on autopilot. if this works, the pace of science just hit fast-forward.
Holy shit...Stanford just built a system that converts research papers into working AI agents. It’s called Paper2Agent, and it literally: • Recreates the method in the paper • Applies it to your own dataset • Answers questions like the author This changes how we do science…
matt pocock nailed it: there are two kinds of prompting. one for building apps (detailed, structured, thought-out) and one for quick tasks (fast, light, gets it done). the trick is knowing which mode you’re in before you start typing.
I have two modes when I'm prompting LLM's. When I'm building an app containing an LLM, I go pretty hard. I use Anthropic's 10-section template. But for one-off prompts to Claude Code, I go down to 4:
robots can now learn by conversation or demonstration. with gemini robotics 1.5, you can literally just talk to the robot or show it what you mean. feels less like coding a machine, more like collaborating with one.
Interacting with robots just got easier with the agentic capabilities of Gemini Robotics 1.5. Talk to the robot or show it things! See how the ER model reads a handwritten list on paper and packs the tools for the job.
just dropped a new tutorial on how to build a hyper-personalized movie recommendation agent using semantic search + sql. it runs on a database of 300k+ movies, and it’s a fun mix of ai and old-school querying. link’s in the next post.
In case you missed it! 💫 Yesterday, I shared a tutorial on building a hyper-personal movie recommendation agent. Built on top of over 300k+ movies, the project combines the power of modern semantic search with SQL. You can also watch it on YouTube (link in the next tweet).
a study just showed you can predict what people will actually buy by asking an llm to *pretend* to be them. it nailed purchase intent with around 90% accuracy. feels like market research just met its simulation theory era.
This paper shows that you can predict actual purchase intent (90% accuracy) by asking an LLM to impersonate a customer with a demographic profile, giving it a product & having it give its impressions, which another AI rates. No fine-tuning or training & beats classic ML methods.
microsoft quietly dropped a free course on building ai agents using their copilot platform. 11 deep-dive lessons, no paywall, high signal. feels like one of those things people will wish they’d jumped on early.
Microsoft quietly dropped a free MCP course, and it’s a total goldmine for anyone building AI agents! 🔥 11 deep-dive lessons showing you how to actually build and deploy production-grade MCP agents. 100% free on YouTube. Links in 🧵 ↓
koyal (by @KoyalAI) is building an agentic ai filmmaker that turns your music, podcasts, or narration into cinematic videos, keeping characters, settings, and storylines consistent. automated direction meets creative control. backed by y combinator.
Koyal (@KoyalAI) is an agentic AI Filmmaking platform that helps you create cinematic videos with consistent characters, settings, and storylines. It turns music, podcasts, and narration into short-form videos and full episodes with tooling to customize any detail.
the 2025 state of ai report just dropped. ai now solves math olympiad problems at gold level and alphazero is literally teaching world champion chess players. we’ve moved from ai competing with humans to ai coaching them. wild shift.
If you want to know what happened in AI this year, the 8th annual State of AI Report is now available. Here are some of the biggest achievements from 2025: - AI achieved math Olympiad gold-level performance for the first time. - AlphaZero taught four world champion chess…
ai video tools like sora aren’t just about faster content, they’re about lowering the walls to creativity. when storytelling is one prompt away, the next wave of filmmakers might not come from hollywood, but from anyone with wi‑fi and an idea.
pydantic ai just dropped v1.0.17, keeping the momentum strong. small update, but every version means better validation and smoother ai workflows. always cool seeing open source move this fast. 🎉
Pydantic AI version 1.0.17 is out! 🎉 github.com/pydantic/pydan…
most ai models can sound smart without actually *being* smart. they don’t think, they just predict what “thinking” looks like. the distinction between performance and intelligence is everything.
The most clear and elegant explanation of what I’ve been thinking about model AI models and why they aren’t intelligent like humans. tomklingenstein.com/language-witho…
simon willison just found that claude’s code interpreter ships with a `/mnt/skills/public/` folder full of python tools for pdf, docx, pptx, and xlsx. basically, claude comes with preloaded skills you can inspect and copy. ai systems are starting to come with starter kits.
I just learned Claude's new code interpreter mode has a /mnt/skills/public/ folder full of prompt instructions and Python utilities for creating and manipulating pdf, docx, pptx, and xlsx files - and you can ask Claude for a copy and learn a TON about working with those formats
maybe intelligence isn’t something we put *into* language, but something that language itself creates. our words don’t just describe how we think, they *are* how we think.
new drop: qwen3-vl cookbooks. a set of hands-on notebooks showing what the model can do, from reading images to coding to full-on agent behavior. available for both api and local use. learn by doing, not guessing.
Introducing Qwen3-VL Cookbooks! 🧑🍳 A curated collection of notebooks showcasing the power of Qwen3-VL—via both local deployment and API—across diverse multimodal use cases: ✅ Thinking with Images ✅ Computer-Use Agent ✅ Multimodal Coding ✅ Omni Recognition ✅ Advanced…
satya nadella just kicked off the first of many nvidia-powered ai system rollouts at microsoft. the scale here isn’t about experiments anymore, it’s infrastructure. the ai arms race just moved another step into production.
Microsoft CEO Satya Nadella offered a glimpse of the "first of many" massive Nvidia AI systems it is rolling out, starting now. techcrunch.com/2025/10/09/whi…
genie 3 just made *time*’s 2025 best inventions list. a world model that can spin up an entire playable world from a single image or text prompt. wild to see generative ai move from pixels and words to living, interactive worlds. huge congrats to the team.
Fantastic to see Genie 3, our state-of-the-art world model, featured in @TIME's 2025 Best Inventions. From a single image or text prompt to an entire playable world, it’s the future of AI and entertainment. So proud of @jparkerholder @shlomifruchter & the team - huge congrats!
amazon just dropped “quick suite,” an agentic ai platform for enterprises. think ai agents that analyze data, summarize insights, and take action across slack, aws, and salesforce. feels like amazon’s move to join the copilot/duet ai race, with a heavy dose of “trust us.”
Amazon just dropped Quick Suite, an agentic AI platform for enterprise ecosystems. Agents can analyze data, summarize insights, and act across Slack, AWS, Salesforce, with full security and privacy! 🔥 Looks super promising.
if every problem was solvable, what would we even dream about next? maybe meaning itself would become the last great puzzle.
United States Trends
- 1. Araujo 96.7K posts
- 2. Estevao 62.2K posts
- 3. Ferran 52.1K posts
- 4. Chelsea 397K posts
- 5. Barcelona 299K posts
- 6. Rashford 20.1K posts
- 7. Kounde 23.5K posts
- 8. Godzilla 22.4K posts
- 9. Rush Hour 4 15.4K posts
- 10. Cucurella 32.6K posts
- 11. #AskTems 3,606 posts
- 12. Sonny Gray 8,438 posts
- 13. #CHEBAR 9,129 posts
- 14. Yamal 73.2K posts
- 15. Happy Thanksgiving 22.7K posts
- 16. Enzo 42.1K posts
- 17. Flick 24.6K posts
- 18. Raising Arizona 1,268 posts
- 19. National Treasure 6,415 posts
- 20. Balde 8,100 posts
Something went wrong.
Something went wrong.