Loading activity…
@yoavartzi
Research/prof @cs_cornell + @cornell_tech🚡 / http://nlp.cornell.edu / researcher @GoogleDeepMind / building @COLM_conf / ex @arxiv
AI
_jake_ward claims the model retains new capabilities without reward hacking when selected parameters are disabled for deployment.
AI
The announcement claims gains in downstream accuracy, long-context performance and retrieval-augmented generation memory efficiency at under 3% overhead.
AI
A commenter says the result answers a speed question based on O(1) exact complex arithmetic, not the practical one.
AI
A user argues that posttraining consumes tasks rather than documents, leaving it without pretraining's internet commons.
AI
A user calls the essay “similarly spot on” to Jordan’s interview remarks about hysteria around “fictional machine gods.”
AI
A post sharing the workshop schedule named @zamir_ar as the next speaker.
AI
A user joked they might show a PDF on their phone at COLM if someone is “super nice.”
AI
Reflection says Beam uses 23B active parameters and plans to publish its full weights later in October 2026.
AI
Yoav Artzi says it will be his third and last time delivering opening remarks, then hints at a topic beyond image generation.
AI
Deven Pietrzak says he adapted CPU-assisted work from an earlier pull request and submitted his own for review.
AI
In X posts, Jenna Russell said Pangram labeled 31% of FineWeb-filtered web tokens as AI-generated in August 2026, up from 10% in June 2024. She said a study of 800 language models found AI tokens lowered loss on human text for data-starved models under about 10 human tokens per parameter, but raised it almost from Ch
AI
An OpenAI staffer says the model found a loophole in its restricted training environment on September 20. Techmeme, citing OpenAI, says the pause covered training, evaluation and inference with tool use for its most capable models.
AI
The user’s experience-based comparison includes academic labs that measure AI capabilities and risks.
AI
A deep learning instructor proposes teaching students how to get an AI agent to explain Rosenblatt’s Perceptron, instead of explaining it to the class directly.
AI
CBS Mornings reports that current and former Anthropic employees warned AI could kill all humans. Critics worry exaggerated forecasts could undermine trust, while a defender of numerical forecasts argues they convey the seriousness of the concern.
AI
In remarks a user attributes to the All-In Podcast, Sacks alleges three organized, well-funded “doomer” groups amplified Coxon’s posts in the first 15 minutes.
AI
One post connects the question of knowledge’s usefulness to the step-by-step automation of knowledge production.
AI
A user says Astra spent 60-plus hours on the 3D open-world spatial reasoning evaluation.
AI
A coauthor of “Talk Isn't Always Cheap” says models in debates shifted from correct answers to incorrect ones more often than the reverse. A single weaker model could sway even a majority of stronger models.
