🔥Telegram İfşa

Excited to share what I have been focusing on this year! Inference-time search to optimize Bayesian surprise pushes us towards long-horizon discovery! Introducing "AutoDS": Autonomous Discovery via Surprisal. "It can not only find the diamond in the rough, but also can rule out

I will be working on RL for drone racing and swarms on stream here/YT/Twitch for the next few hours. Goal is a ~100k param multitask model that we can deploy on real hardware

Infinite Wiki ⁕ Every word is a hyperlink. Every description is generated in real-time (in ~1 second) ⁕ Runs on Gemini 2.5 Flash Lite. ASCII diagrams using 2.5 Flash

Jack Dorsey has recently released two apps in less than a week. How? He's riding the vibe coding wave and using an AI tool called Goose to crank out new apps for messaging and tracking sun exposure.

Today, we at OpenAI achieved a milestone that many considered years away: gold medal-level performance on the 2025 IMO with a general reasoning LLM—under the same time limits as humans, without tools. As remarkable as that sounds, it’s even more significant than the headline 🧵

Their bet allowed for formal math AI systems (like AlphaProof). In 2022, almost nobody thought an LLM could be IMO gold level by 2025.

With the rising demand for AI model inference, optimizing these models is the key to real-world deployments. Techniques like Pruning, Distillation, Quantization, efficient KV caching, MQA/GQA/MLA attention, speculative decoding, dynamic batching, MoE, and test-time compute

Noam tends to not exaggerate.

How can we combine retrieval and reasoning for better deep research? ▪️ Reasoning → RAG: Use reasoning to improve parts of the RAG pipeline. ▪️ RAG → Reasoning: Use retrieved facts to make the model’s reasoning more grounded and accurate. ▪️ Synergized RAG and Reasoning (the

Been cooking something up with Claude

we achieved gold medal level performance on the 2025 IMO competition with a general-purpose reasoning system! to emphasize, this is an LLM doing math and not a specific formal math system; it is part of our main push towards general intelligence. when we first started openai,

openAI Noam Brown confirmed 2 key points: 1) the model that reached IMO gold-level performance is general-purpose — not fine-tuned for math, and achieved it without using tools 2) openAI developed new techniques that make LLMs a lot better could it be a new RL breakthrough or

🚨 BREAKING: Detailed list of all 44 people in Meta's Superintelligence team. — 50% from China — 75% have PhDs, 70% Researchers — 40% from OpenAI, 20% DeepMind, 15% Scale — 20% L8+ level — 75% 1st gen immigrants Each of these people are likely getting paid $10-$100M/yr.

Lucy (1.7B): A Qwen3-based agentic web search model trained purely with RL (no SFT). Optimized for mobile and CPU-only use. - Agentic search via Serper (Google) + Crawl4AI - Machine-generated task vectors guide reasoning - Outperforms DeepSeek-v3 on SimpleQA (MCP benchmark) -

Tip: You can run multiple code agents in parallel. Works well with Gemini CLI, Claude Code + git-worktree.

Gemini Code Assist now shares the same technology with Gemini CLI! In VS Code, place any prompt into the chat window using agent mode, and Code Assist will work to write tests, fix errors, build out features, or migrate your code → goo.gle/4o5eEp4

Google Code Assist now has the Agent Mode. It can build out features, write tests, fix errors, and even migrate your code. Cursor like experience but 100% free. Let that sink in.

Distributional Reinforcement Learning - Published by MIT in 2023 - One of the best book in Reinforcement Learning Link in comments 😎 👇

Lizard, a linearization framework that transforms pretrained Transformer-based Large Language Models (LLMs) into flexible, subquadratic architectures for infinite-context generation. Lizard shows a transformer can keep almost all its smarts while replacing softmax with cheaper

New Essay by @BlaiseAguera (Google): “AI Is Evolving — And Changing Our Understanding Of Intelligence” Advances in AI are making us reconsider what intelligence is and giving us clues to unlocking AI’s full potential. noemamag.com/ai-is-evolving…