- Replying to @OrdinaryGamersCompletely insane how they missed the opportunity with rdr2 single player dlcs. Single player has an enormous potential.
- Traditional Neural Networks are overrated. Transformer is irrelevant when we have a small set of unique examples. And no, brute-forcing with discrete program search is not the way too. It's so hilarious how easy, simple and yet elegant the actual approach is.
- DeepSeek R1 vs Flash-thinking-01-21 2.0 in my high-leverage reasoning benchmark to test agency in test time models. R1: 53% Flash-thinking: 23% o1: 30% R1 is way better in multi step reasoning The R1 is the most thoughtful of them all.
- The last question for LLMs: "Sort the numbers from highest to lowest: 9.1, 9.8, 9.11, 9.9, 9.12" o1-preview: failed o1-mini: failed gpt-4o-latest: failed sonnet-3.5: failed llama-3.2-1B: success
- OpenAI released a Music Video with Sora it's puzzling i dont think ive ever seen anything quite like this before
00:00 - It's absolutely insane, R1 thought for 10k+ tokens on aime problem I gave it and R1 solved it correctly. If they release open source we are unfathomably back
- multi-trillion-dollar industry hangs on the gpt2-chatbot: if it's gpt-4.5/5 model: it's over if it's a gpt-2 sized model: we are unfathomably back
- DeepMind will be first to achieve AGI.With @Harvard, we built a ‘virtual rodent’ powered by AI to help us better understand how the brain controls movement. 🧠 With deep RL, it learned to operate a biomechanically accurate rat model - allowing us to compare real & virtual neural activity. → dpmd.ai/3RobU7e
00:00 - Forget about GPT-5 Here is SOTA AI model for predicting the structure and interactions of ALL life’s MOLECULES. WE ARE BACK!Announcing AlphaFold 3: our state-of-the-art AI model for predicting the structure and interactions of all life’s molecules. 🧬 Here’s how we built it with @IsomorphicLabs and what it means for biology. 🧵 dpmd.ai/3URDiNo
00:00
