“Sharpness-Aware Pretraining Mitigates Catastrophic Forgetting” Better base models don’t always become better fine-tuned models, because post-training can overwrite pretrained capabilities. So this paper pretrain models into flatter loss regions, so later updates from
LLMS
-
Manifold Steering for Neural Network Representation and Behavior
By
–
"Manifold Steering Reveals the Shared Geometry of Neural Network Representation and Behavior" Most activation steering treats concepts like directions, but model representations are often curved, so straight line steering can cut through unnatural internal states. The key idea
-
Technical clarification on LLM data processing and regurgitation mechanisms
By
–
Since Hinton has actually replied let me clarify some things – LLMS don’t *always* regurgitate
– LLMs don’t literally store full texts
– but given the mechanisms that they use they do sometimes regurgitate long passes through their process of resynthesis
– the evidence for that -
The Evolution and Challenges of AI Coding Agents
By
–
The question has moved past whether coding agents can produce code that works. The harder question is whether they can complete real software work safely, measurably, and repeatedly — and how we supervise it. Great piece by @realjustinbauer on the evolution of coding agents and
-
Analyzing the generative mechanics and hallucination patterns of LLMs
By
–
they break things into bits. then they reconstitute them often the reconstitution includes verbatim repetition even the hallucinations are often combinations of real bits but from different people, articles that shouldn’t be combined all that has the flavor of pastische
-
On the nature of LLM memorization and data compression
By
–
that’s why i always say to a first approximation and when given space go into why’s it not just memorization . but some of it is effectively memorization, in same extended sense as a JPEG is.
-
Compilation of 25 Powerful AI Skills for ChatGPT, Claude, and Gemini
By
–
Anthropic explicó en este vídeo por qué las Skills son el futuro de la productividad con IA.
— Nico (@nicos_ai) 11 mai 2026
Vídeo subtitulado al español.
En este artículo recopilé las 25 Skills más potentes para usar con ChatGPT, Claude y Gemini.
Muy recomendable 👇https://t.co/GYvJyuLdYn pic.twitter.com/HR00uSP07bAnthropic explained in this video why Skills are the future of productivity with AI. subtitled in Spanish. In this article, I compiled the 25 most powerful Skills to use with ChatGPT, Claude, and Gemini. Highly recommended
-
Using LLM CLI tools for executable scripts in shebang lines
By
–
New TIL: I figured out how to use my LLM CLI tool in a shebang line, which means you can write executable scripts in English, or hook up more complex scripts with a snippet of YAML template
-
New Memory feature update allows context transfer between AI chats
By
–
new Memory update ✨ pt.6
— Character.AI (@character_ai) 11 mai 2026
🆕 Copy Memory – start a new chat without losing the plot🔐
bring Story Memory and Facts along to a new chat, or leave it all behind. up to you.
more memory updates coming soon (•◡-)♡ https://t.co/9Y9aQvCJQ8 pic.twitter.com/WbtSnG4XCpnew Memory update pt.6 Copy Memory – start a new chat without losing the plot bring Story Memory and Facts along to a new chat, or leave it all behind. up to you. more memory updates coming soon (•◡-)♡