*AI does something wild* Skeptics: yawn. so what. it was trained on everything AI read every book ever written... therefore... it can't do anything interesting?? Imagine meeting a HUMAN that read every book?
Sorry, but anybody who still says "LLMs just predict text" is embarrassing themselves at this point It is, as OpenAI's roon says, "categorically wrong"
APEX is a first-of-its-kind benchmark that evaluates AI models based on their ability to perform economically valuable knowledge work. At the moment, GPT-5 is leading all metrics. But remember, this is the worst it will ever be.
AI has its PhD and now it’s on the job market. Introducing the AI Productivity Index (APEX), a benchmark that measures how well we’ve automated the most valuable industries in the world. Most benchmarks study abstract capabilities. APEX evaluates model performance on real…
Verifier with NO trained parameters?🤯 Even outperform GPT-4o in terms of verification accuracy. We at Tencent AI Lab introduce CLUE - 🕵️ a verifier based on clustering where successful vs. failed reasoning creates separate hidden states. Paper link: huggingface.co/papers/2510.01…
Google just dropped a new Generative AI Python library for SQL Databases. Introducing Google GenAI Toolbox. This is what you need to know:
Sam Altman says technology is outpacing our wisdom, which leaves society unbalanced "this was the year AI got smarter than us" Life goes on, but something fundamental has changed The future is wide open: digital immortality, disease cures, no one knows how far it goes
much more convinced after getting my own results: LoRA with rank=1 learns (and generalizes) as well as full-tuning while saving 43% vRAM usage! allows me to RL bigger models with limited resources😆 script: github.com/sail-sg/oat/bl…
LoRA makes fine-tuning more accessible, but it's unclear how it compares to full fine-tuning. We find that the performance often matches closely---more often than you might expect. In our latest Connectionism post, we share our experimental results and recommendations for LoRA.…
🚨BREAKING: Creative jobs could be obsolete by 2026 AI can now design, test, and optimize 1000s of catalog ads without a human team. This founder raised $10M to build the platform killing UGC ads 🧵
We found that visual foundation encoder can be aligned to serve as tokenizers for latent diffusion models in image generation! Our new paper introduces a new tokenizer training paradigm that produces a semantically rich latent space, improving diffusion model performance🚀🚀.
To the extent that modern post-training as a paradigm violates the bitter lesson by being human engineered, it is evidence against a purist view of the bitter lesson. Not the other way around. AI is not abstract. It’s a highly human-centered, utility-driven engineering domain.
Finally had a chance to listen through this pod with Sutton, which was interesting and amusing. As background, Sutton's "The Bitter Lesson" has become a bit of biblical text in frontier LLM circles. Researchers routinely talk about and ask whether this or that approach or idea…
LLMs aren't a dead end or hitting a wall their issues are solvable, but will take longer than many expect the real gap is hype vs. reality the dot-com bubble proved the internet thesis was right, just early and overbuilt today feels similar: same pattern, bigger and faster
Aravind Srinivas argues that generic financial advice is now replicable by AI faster and more thoroughly than any human. Models like Perplexity can read every analyst report, summarize market trends, and tailor strategies in real time. But human advisors remain valuable if they…
I am very excited to announce that the third CPAL conference: cpal.cc is to be held at Tübingen, Germany in March 2026 (after Hong Kong and Stanford). Complementary to the arguably too many mega conferences on machine intelligence, CPAL aims to be focused and…
A secret I believe that most don’t is this: Computers are becoming more trusted than humans. > Humans can lie, cheat, or change their minds. > Computers, when governed by immutable code, can’t. >The longer they prove this reliability, the more we hand them our trust. That’s…
another day, another agent introducing "Caesar" — an AI agent that uses your computer like a human what it does: - automates tasks across desktop, mobile, and web - runs on any device and any OS - handles software testing beyond the web No APIs. No integrations.
Introducing ⚡ Caesr by @ask_ui The AI agent that operates your computer like a human. Clicks, types, navigates — across desktop, mobile, and web. No APIs. No integrations. Just tell it what to do.
Lovable cloud just dropped. Here's what it means: • Add API's, Databases or even Stripe directly into your build in Lovable • One-click OpenAI API • Baked in MCP's • Custom backends No more work around, just unlimited possibilities!
@DynaRobotics successfully zero-shot folded a new @corl_conf shirt -- even with the sleeve tucked in awkwardly! Amazing stuff. @JasonMa2020
Everyone keeps talking about this demo, if anyone won corl it was these guys
It blows my mind that you can do this now! I opened @get_mocha and built an entire application in less than 30 minutes: • the frontend • the backend • with a database • with Stripe integration I used to charge clients 5 figures to do all of this, and I finished it today in…
From replatforming to edge-powered AI 🚀 Join VSCO CTO Chris Haire live for a behind-the-scenes look at VSCO’s modernization journey—and how they built Canvas, an AI tool scaled at the edge. Register now: cfl.re/48ERv7q
“Sir, Dario just dropped claude 4.5 and it beats GPT-5 in coding, agentic tasks, and computer use AND deepseek dropped new model with 10x cheaper inference 50%+ cheaper API costs…”
New Microsoft Research paper stress-tests GPT-5, Gemini 2.5, GPT-4o & others on medical benchmarks and uncovers deep fragilities beneath the leaderboard wins. Key Findings: — Models still guess correctly even without images — Reasoning is often fabricated or wrong — Simple…
United States 趨勢
- 1. Baker 30.5K posts
- 2. #BNBdip N/A
- 3. Cowboys 72.6K posts
- 4. Packers 28.7K posts
- 5. 49ers 31.3K posts
- 6. Flacco 11.2K posts
- 7. Fred Warner 10.5K posts
- 8. Panthers 73.8K posts
- 9. Cam Ward 2,744 posts
- 10. Bucs 9,801 posts
- 11. Niners 4,982 posts
- 12. Zac Taylor 2,854 posts
- 13. Tez Johnson 2,863 posts
- 14. Titans 22.8K posts
- 15. #FTTB 3,972 posts
- 16. Ja'Marr Chase 2,382 posts
- 17. Browns 64.8K posts
- 18. #Bengals 3,003 posts
- 19. Yoshi 32K posts
- 20. #GoPackGo 3,563 posts
Something went wrong.
Something went wrong.