Great to catch up with @akothari yesterday in Miami (Is Miami back btw?). @NotionHQ is killing it with agents and just rolled out their first open weight models with Minimax. In my (biased) opinion, applied AI startups can’t solely rely on proprietary APIs who end up being
LLMS
-

CUDA Agent: Large-Scale Agentic RL for Kernel Generation
By
–
CUDA Agent | Large-Scale Agentic RL for CUDA Kernel Generation https://
buff.ly/13zVaiQ
#AI #MachineLearning #DeepLearning #LLMs #DataScience -
Agent 4 Deep Dive: Everything You Need to Know
By
–
Agent 4 Deep Dive – Everything You Need to Know https://
x.com/i/broadcasts/1
pKkOyrnkYYKj
… -

Advisory Jury in Musk-OpenAI Trial Focuses on Liability, Not Damages
By
–

What is the significance of microbes in tumors, a field mired in controversy? https://
cell.com/cancer-cell/ab
stract/S1535-6108(26)00109-1
… -
Token Generation Speed Comparison: OSS and Nemotron Models Performance
By
–
Yeah. Not quite as bad for me, but:
gpt-oss-120b: 42.22 tokens/sec
nemotron-3-super: 20.43 tokens/sec on the DGX Spark. But this is ollama and might be an implementation issue. I have yet to try the Nvidia-optimized llama.cpp version (they had one for Nemotron Nano back then) -

Live Demo of Agent 4 Building a Realtime Chat App on Replit
By
–
We're going LIVE in less than an hour to show you what Agent 4 can do! Building a realtime chat app that works across web and mobile, all inside one Replit workspace. Starting on the new Agent Canvas, then building out a full web and mobile app from scratch. 9AM PT on YouTube,
-
Frontier Models: Cleaning Up Lower-Generation AI Output
By
–
The biggest usecase for advancing the frontier models is to clean up slop written by frontier(n-1) models! Arguably, skilled programmers writing building on quality code could manage just fine on prior generations of models by just improving tools, harnesses, skills, etc. With
-

Barbell Attention Architecture: Comprehensive Guide to Advanced Transformers
By
–
Barbell Attention is All You Need?
-
AI’s Current Bottleneck: Pattern Memorization Requires Human Guidance
By
–
The bottleneck of current AI is simple: the techniques we use are still predicated on pattern memorization and retrieval, and thus they need *someone* to tell them which patterns to memorize (training data, RL envs…) That role cannot yet be played by AI in a truly open-ended
-
Codex Model Efficiency Falls Short of OpenAI Expectations
By
–
Interesting that it wasn't more efficient than 5.3 codex, like OpenAI said it should be
