Reinforcing Recursive Language Models Can a 4B model learn to recursively call itself to answer hard long-context questions? We RL fine-tuned a small model to behave as a native RLM. On evidence selection across scientific papers, our 4B RLM matches Sonnet 4.6 in quality
RESEARCH
-
Exchange with OpenAI’s Roon on superintelligent AI and organizational challenges
By
–
Had an interesting exchange with roon of OpenAI last night over whether super intelligent AI would actually be able to navigate organizational challenges.
-
Anthropic’s Internal Alignment as a Driver for AI Development Pace
By
–
One secret to @AnthropicAI's blistering pace: strong internal mission-alignment https://t.co/z6RGBklQqw pic.twitter.com/NouZuhDbh1
— Lenny Rachitsky (@lennysan) 12 mai 2026One secret to @AnthropicAI
's blistering pace: strong internal mission-alignment -
Google DeepMind introduces AI Co-Mathematician agent for research
By
–
NEW paper from Google DeepMind. (bookmark it) AI Co-Mathematician is an agentic research workbench for mathematicians, and it just hit 48% on FrontierMath Tier 4, a new high score among AI systems evaluated. The system is an asynchronous, stateful environment that supports
-
Symbolic Learning as a Scalable Alternative to Neural Networks
By
–
Symbolic learning is not a replacement for coding agents, it's a replacement for gradient descent & NNs: a low-level, completely general, extremely scalable new learning substrate.
-
AGI ALPHA: A Scalable Substrate for Intelligence Organizations Manuscript Published
By
–
AGI ALPHA: A Scalable Substrate for Intelligence Organizations Vincent Boucher, President of http://
MONTREAL.AI and http://
QUEBEC.AI Paper: https://
github.com/MontrealAI/agi
alpha-first-real-loop/blob/main/docs/manuscript/AGI_ALPHA_Unified_Publication_Final.pdf
… #AGIAlpha #QuebecAI #SovereignAI -
Gary Marcus’s Warnings on AI Generalization Proven True
By
–
“Marcus's repeated warnings about the "wall of generalization" since 1998 have once again been proven true.”
-
GPT-5.5 evaluation errors revealed in FrontierMath benchmark
By
–
The irony. Using AI (GPT 5.5), errors have been identified that would affect 1/3 of the problems in the FrontierMath benchmark Tiers 1-4. This could represent a shift in the evaluations of AI's mathematical capabilities, which may be underestimated.
-
Multi-Agent Synergy for Scaling Test-Time Compute
By
–
TMAS Scaling Test-Time Compute via Multi-Agent Synergy
-
Rebellious Student: Reversing Teacher Signals
By
–
Rebellious Student Reversing Teacher Signals for Reasoning Exploration with Self-Distilled RLVR