3. DeepDive Builds a stronger web-browsing deep search agent by pairing two ingredients: automatically synthesized, hard-to-find questions from knowledge graphs and end-to-end multi-turn RL that teaches the model how to reason, search, and stop.
LLMS
-
Traditional RL Researchers and Their Opposition to Supervised Learning
By
–
Traditional RL folks were for some weird reason against supervised learning and LLMs. Have they learned the bitter lesson?
-
MAI1 preview: New young team behind major AI breakthroughs
By
–
Thanks for asking: MAI1-preview, MAI voice. Most of the team is less than a year old, O(100) and made of the people that just recently helped ship MovieGen, Imagen3, Veo, Gemini, Llama, NotebookLM, Genie, etc, …. Like the 90s La-di-da, da-di-da
Di-dai-dai-da
More than words -
Young team achieves rapid progress in LLM and voice technology
By
–
Maybe not so. The team has existed for less than a year and has already landed a LLM preview in lm arena and several voice products – I believe the most expressive, efficient and highest quality voice generation in history. We’re improving fast, and whoever applies will make many
-
LLM Plugins: Making Models Available Through Python Library
By
–
Yes, any plugin for LLM also makes those models available to the Python library Here's a bit of the documentation that covers that https://
llm.datasette.io/en/stable/pyth
on-api.html#models-from-plugins
… -
Single-step RL in Language Models: Differentiability and Token Generation
By
–
+1 to @TacoCohen on differentiability. Moreover it is important to remember that most RL for LMs involves only one step of RL where the state (question, instruction) is provided by the environment and where RL must generate a single sequence of tokens (1 action). This results in
-
Major AI Companies Accelerating Release Cycles in 2025
By
–
What we see is that the major AI companies have significantly increased the number of publications in recent months. From Grok 1 to Grok 1.5, it took about a year, followed by another year-long break. But in 2025 alone, Grok 2, 3, and 4 were released. At OpenAI, you can see
-
GPT-4 Powered Malware MalTerminal Raises Security Concerns
By
–
GPT-4 Powered Malware Is Here: Researchers Call MalTerminal a Wake-Up Call https://
search.app/dRNQC #malware #GPT #LLMs #GenerativeAI #AI #ArtificialIntelligence #GenAI #security #CyberSecurity #CyberSec @lexfridman @KirkDBorne @Ronald_vanLoon @erikbryn @antgrasso @sallyeaves -
Code Hidden in LLM Layer Activations Demands Apology
By
–
The code was written at layers 22-30 and is stored in the value activations you just can’t read it. I think you owe the LLM an apology.