This MIT paper explains why telling models to “think harder” keeps failing in practice. The core argument is simple and uncomfortable: reasoning doesn’t break because models lack knowledge. It breaks because reasoning itself becomes unstable when it goes on too long. The
RESEARCH
-
AI Cannot Learn Specific Knowledge Without Verifiable Answers
By
–
Specific knowledge doesn’t have verifiably correct answers – so it’s not something AI can be trained for, either.
-
Simple Systems Scaled Up: Understanding AI Architecture
By
–
In every working complex system, you see something simple scaled up, and AI is no different.
-
Datasets: Accounting for Size and Quality Metrics
By
–
Is this # datasets without accounting for size or quality? https://t.co/7dv2Lv6EBj
— Ken Goldberg (@Ken_Goldberg) 24 décembre 2025Is this # datasets without accounting for size or quality?
-
Papers Rejected for Legal Code Instead of Greek Pseudocode
By
–
Would papers get rejected if they used legal code with variable names instead of Greek laden pseudo code?
-

Poetiq System Surpasses Human Baseline on ARC-AGI-2 Benchmark
By
–


BREAKING : Poetiq system with GPT-5.2 X-High as a base, broke through human baseline on ARC-AGI-2 benchmark. From 65% to 75% in a month
-

First CAIS Conference on AI and Agentic Systems May 2026
By
–
Happy to support this effort together with many great people. Consider submitting your work to this very first conference (CAIS) on AI & Agentic Systems! Submission deadline: Feb 27, conference May 26-29, 2026 in San Jose, CA. https://
caisconf.org. -

Text-Graph Complementarity in NLP: R²-CoD Analysis Framework
By
–
💡Text + graphs have powered NLP tasks for a long time. But key questions remain: How do their representations relate to each other during learning? How does that relationship impact task performance? Under what conditions is integrating text and graph more beneficial? In our paper, "R²-CoD: Understanding Text-Graph Complementarity in Relational Reasoning via Knowledge Co-Distillation", we take an analysis-driven perspective and uncover a spectrum of behaviors from complementarity → partial alignment → full alignment. 📜arxiv.org/abs/2508.01475 🧵[1/8]
-
Sam Altman Says Total Memory Will Be AI’s True Breakthrough
By
–
Sam Altman affirme que la véritable percée de l’IA ne viendra pas d’un meilleur raisonnement, mais d’une mémoire totale.
— VISION IA (@vision_ia) 23 décembre 2025
Une IA sera capable de se souvenir de chaque conversation, email et document tout au long de la vie d’une personne, en identifiant des schémas et des… pic.twitter.com/QUhmpHGV2OSam Altman affirme que la véritable percée de l’IA ne viendra pas d’un meilleur raisonnement, mais d’une mémoire totale. Une IA sera capable de se souvenir de chaque conversation, email et document tout au long de la vie d’une personne, en identifiant des schémas et des
-

GLM-4.7 Foundation Model Achieves Record-Breaking Reasoning and Coding Scores
By
–
Now on Poe: GLM-4.7 This next-generation foundation model, powered by @novita_labs on Poe, posts record‑breaking scores across several major reasoning and coding benchmarks. (1/2)
