Another by Yann LeCun! “The Spike, the Sparse and the Sink” This paper shows that massive activations and attention sinks come from the same pre-norm Transformer pipeline rather than being separate anomalies. Early SwiGLU blocks act like directional quadratic amplifiers,
RESEARCH
-

Agentic Data Science: Implementing AI Agents for Data Analysis
By
–
How to Do Agentic Data Science https://
buff.ly/p03xjBG
#AI #MachineLearning #DeepLearning #LLMs #DataScience -
AI Research Labs Affected by OAuth Outage, Intelligence Brownouts Concern
By
–
My autoresearch labs got wiped out in the oauth outage. Have to think through failovers. Intelligence brownouts will be interesting – the planet losing IQ points when frontier AI stutters.
-
PostTrainBench v1.0: Evaluating AI Agents for Model Post-Training
By
–
Excited to release PostTrainBench v1.0!
— Karina Nguyen (@karinanguyen) 11 mars 2026
This benchmark evaluates the ability of frontier AI agents to post-train language models in a simplified setting.
We believe this is a first step toward tracking progress in recursive self-improvement 🧵: pic.twitter.com/ELymwJqVP1Excited to release PostTrainBench v1.0! This benchmark evaluates the ability of frontier AI agents to post-train language models in a simplified setting. We believe this is a first step toward tracking progress in recursive self-improvement 🧵:
-

WMPO enables robots to learn complex tasks without real-world errors
By
–
What if robots could learn complex manipulation tasks and self-correct, all without costly real-world errors? Researchers from Hong Kong University of Science and Technology and ByteDance Seed present WMPO (World-Model-based Policy Optimization). This new framework lets
-
Brockman’s statements on deceiving Musk and OpenAI’s early contributions
By
–
More on the study results and interpretationhttps://t.co/IAeO7zFL5U
— Eric Topol (@EricTopol) 11 mars 2026More on the study results and interpretation
-
Multilingual LLM Pattern Coverage: Feasibility and Validation Challenges
By
–
Multilingual can't possible work, do you really plan to identity patterns in every human language understood by an LLM? L How would you tell if you've covered them all, especially given the range of available LLMs?
-

Musk-OpenAI Trial: Mission vs. Profit Debate
By
–
In their paper today, the authors assert that LLMs are "homogenizing" human thought and expression https://
cell.com/trends/cogniti
ve-sciences/fulltext/S1364-6613(26)00003-3
… -
AI and Robotics Breakthroughs: One Year of Rapid Innovation
By
–
I break down stories like every day this in my free newsletter. Keep up with the latest in AI/Robotics in 5 min a day: therundown.ai/subscribe Source: SWARM Biotactics linkedin.com/posts/s-wilhelm…
→ View original post on X — @rowancheung, 2026-03-11 16:15 UTC
