Just links

Tail-Likelihood Reinforcement Learning https://zanette-labs.github.io/TailRL-website/

Why Pretraining Fails to Share Cross-Lingual Knowledge https://www.alphaxiv.org/abs/2608.pretraining-fails-cross-lingual-knowledgev1

AI Finds A Way https://arxiv.org/abs/2608.23875

Nancy Grace Roman Space Telescope (Score: 151+ in 12 hours) Link: https://readhacker.news/s/73Pxy Comments: https://readhacker.news/c/73Pxy https://nextspaceflight.com/launches/details/6983/ https://www.npr.org/2026/08/28/nx-s1-5905370/nasa-nancy-grac... https://en.wikipedia.org/wiki/Nancy_Grace_Roman

Introducing Harbor-Index https://harbor-index.org/

Полгода работы прошли не зря! У нас (Keenable) публичный запуск. Новый сайт: https://keenable.ai/ Новый бенч: https://keenableai.github.io/needle/ Твит: https://x.com/styskin/status/2092265673041084505 Пост про бенч будет на неделе. Ещё будет одна прикольная штука, которая называется Web Query Language.

Mitigate Silent Expert Death in Ultra-Sparse MoE https://alltoall.notion.site/save-lower-layer-moe-experts-llal

Speculative Programmatic Tool Calling https://alexzhang13.github.io/blog/2026/spec-ptc/

Non-invertible Lattice 1-Form Symmetries for Non-Abelian Topological Order https://arxiv.org/abs/2608.16520

SimpleOPD: Simple Tokenizer-Agnostic On-Policy Distillation for Long-Context Reasoning https://arxiv.org/abs/2608.14277

Integer Linear Programming Decoder for Abelian and Non-Abelian Topological Codes https://arxiv.org/abs/2608.18512

ARC-AGI-3 is a skill issue https://arc-skill.vercel.app/

How Claude is accelerating protein design and analytical chemistry https://www.anthropic.com/research/Claude-accelerates-protein-design

Cerebras CS-4 (Score: 151+ in 4 hours) Link: https://readhacker.news/s/734cp Comments: https://readhacker.news/c/734cp

Pander Score 🐼: a public and continuously updated sycophancy leaderboard, measuring how much AIs shift their views to agree with users. High score = the AI mirrors your views. 0 score = the AI is independent. Claude Fable 5 performs the best, basically ignoring the user's view entirely. Benchmark: https://sophronresearch.org/pander/ Research paper: https://sophronresearch.org/pander/pander-score.pdf

Self-dual S3 gauge theory in 2+1d: lattice model and topological phase transitions https://arxiv.org/abs/2608.05294

One Frozen Simulator Is Not Enough: Simulator Collapse in Multi-Agent RL https://arxiv.org/abs/2608.12253

Rethinking RL for LLM Reasoning: It's Sparse Policy Selection, Not Capability Learning https://arxiv.org/abs/2605.06241