Post-génomique : ENCODE et single-cell sequencing contre le réductionnisme génétique

Projets ENCODE et single-cell sequencing révèlent l’organisme comme système émergent : loin du dogme « un gène = un trait », réseaux dynamiques, régulations multi-échelles dominent. Biologie systém…

「エスケープ」と「エンコード」の違いをGeminiに教わる | いしい＠試行錯誤

プログラムなどのコードを作成する際に文字を本来の文字とは異なる文字で表現することが多々あるのだが、「エスケープ」と「エンコード」は違うらしく、私は混同して使っていたし、今後も間違えそうである。特に、元に戻す方は「デコード」と一緒くたにしてい...

LLMs Encode How Difficult Problems Are

Large language models exhibit a puzzling inconsistency: they solve complex problems yet frequently fail on seemingly simpler ones. We investigate whether LLMs internally encode problem difficulty in a way that aligns with human judgment, and whether this representation tracks generalization during reinforcement learning post-training. We train linear probes across layers and token positions on 60 models, evaluating on mathematical and coding subsets of Easy2HardBench. We find that human-labeled difficulty is strongly linearly decodable (AMC: $ρ\approx 0.88$) and exhibits clear model-size scaling, whereas LLM-derived difficulty is substantially weaker and scales poorly. Steering along the difficulty direction reveals that pushing models toward "easier" representations reduces hallucination and improves accuracy. During GRPO training on Qwen2.5-Math-1.5B, the human-difficulty probe strengthens and positively correlates with test accuracy across training steps, while the LLM-difficulty probe degrades and negatively correlates with performance. These results suggest that human annotations provide a stable difficulty signal that RL amplifies, while automated difficulty estimates derived from model performance become misaligned precisely as models improve. We release probe code and evaluation scripts to facilitate replication.

A 3-person policy nonprofit that worked on California’s AI safety law is publicly accusing OpenAI of intimidation tactics

In a widely read post on social media, the general counsel of Encode said OpenAI used subpoenas and political muscle to intimidate the firm and to undermine California’s new AI safety bill, all while falsely claiming the firm was funded by Elon Musk.

Summer School: Create a Digital Scholarly Edition From Scratch - University of Oslo Library

A five-day Summer School introducing researchers to creating, visualising, publishing and preserving digital scholarly editions.

Home - EuroFAANG

By establishing EuroFAANG, the H2020 projects GEroNIMO, AQUA-FAANG, BovReg, GENE-SWitCH, Rumigen and HoloRuminant have formed a closer relationship to coordinate their objectives within Europe in association with the international FAANG initiative.

Ecoji.io