Widening the Bottleneck

This week: Haiku to Opus in just 10 bits, QKV variants, next-latent prediction transformers, ai moral status, motivated reasoning, emotion concepts, AI designed radio chips, plan A

The Intelligence Organizes Itself

This week: AutoScientists, reward hacking, evolving skill-structure jailbreak, scientific conclusions, AI negotiations, can I buy your KV cache, the office altar.

Where Control Lives

This week: Refusal trajectories, runtime harnesses, skillopt, ubi, panel data

How Do We Study What We Fear?

This week: Model organisms, measuring goal-level contributions, preference for explainable AI, do transformers need three projections, SEGA, AI isn’t management, predictive data debugging

Learning what data to learn from

This week: Agent Island, general intelligence, infinite transformers, curvature

The Inner Organization

This week: LLM safety from within, faithful reasoning, skills to talent, kanbots, don’t paste the ai

What The 0.1% Knows

This week: Agentic world modelling, efficient online memory, who wins polymarket, deepseek-v4, hermes tools

Pandora Opens It

This week: Continual learning trajectories, Pandora’s regret, Dflash, co-packaged optics supply chains, some pytorch

Memory Changes Everything

This week: The memory curse, language style matching, trust scaling, EMO, manifolds