Real reasoning beats CoT hacks.
MACHINE LEARNING
-
Agentic RL Breakthrough: Solving Long Rollout Collapse
By
–
HUGE if you care about agentic RL most agent RL setups collapse under long rollouts either you spam tools endlessly or CoT devolves into junk in this paper, they fix it: > memory overflow detection > single-turn rollout format > length-normalized REINFORCE RECOMMENDED READ
-
Free AI Agent Mastery Guide: Architecture, Use Cases, Tools, Templates
By
–
Everyone’s talking about AI agents. But nobody's explaining how to build them properly. So I made a free AI Agent Mastery Guide for you: → Agent architecture
→ Real use cases
→ Automation tools
→ Prompt templates Like + comment “Guide” and I’ll send it your way. (Must be -
LLMs: The illusion of reliability, not just confident fiction.
By
–
how crazy it is that people expect LLMs to “just admit uncertainty.” they don’t know truth – they only predict the next likely word, and when data is missing, they fill gaps with confident fiction. the real danger isn’t hallucination, it’s the illusion of reliability.
-
Simplifying RL Training: Beyond Enterprise Dependencies
By
–
Luckily, codebases like NanoGPT or the speedrun completely reset people's expectations. But I couldn't find the same for RL; just training Qwen3-4b locally requires hundreds of entreprise-y packages, many of which glitch out when you get slightly off the beaten path…
-
Berkeley EECS launches graduate application assistance program
By
–
Are you applying to grad school this fall? The Equal Access to Application Assistance (EAAA) program for @Berkeley_EECS is now accepting applications! Any PhD applicant to @Berkeley_EECS can submit their application for feedback by Oct. 5th 11:59PM PST. EAAA is a
-
Human Expertise and Iteration Drive AI Model Improvements
By
–
“Experts in the loop + tight iteration drove real gains for Dragon” (via Asha Sharma on @lennysan
). Very aligned with how we build at Snorkel. -
Thinking Machines’ Diverse AI Research: From Kernels to Prompts
By
–
Wow, Thinking Machines’ research is so varied. It ranges all the way from kernel numerics to prompt engineering. I can’t imagine needing anything more to solve AI.
-
Do Humans Learn Through Adversarial Methods?
By
–
humans learn adversarially? this is the first im hearing of this