can’t believe people assume that success on highly verifiable problems in math (where we don’t even know how many tests were performed and how many might have failed) iautomatically generalize to everything else when there is not a shred of evidence that they do.
MACHINE LEARNING
-
Karpathy launches team for Claude-powered recursive self-improvement
By
–
Karpathy will help launch a new team focused on using Claude itself to accelerate pretraining research. Its team is focused recursive self improvement.
-
SenseNova-U1-A3B-MoT Model Weights Open-Sourced
By
–
6/ The full Technical Report is now out, their most detailed model disclosure yet. SenseNova-U1-A3B-MoT (38B-A3B MoE) weights are now open-sourced. You can check out the report or try the tools at the links below. Try it here: https://
unify.light-ai.top/login?next=
home
… Technical Report: -
Open-sourcing Lite series with 8B and MoE models plus fast distilled LoRA
By
–
5/ They are open-sourcing the Lite series in two sizes, an 8B dense model and an A3B mixture-of-experts version. The product has some exciting new updates. An 8-step distilled LoRA now open-sourced: inference cut from 23s to 2s on H100 (100 NFE -> 8 NFE).
ComfyUI is now -
Mix-Quant: Quantized Prefilling and Decoding for Agentic LLMs
By
–
Mix-Quant Quantized Prefilling, Precise Decoding for Agentic LLMs
-
Auth Proxy: Controlling Agent Behavior Boundaries
By
–
Introducing the sandbox Auth Proxy: A way to control the boundary between agent-generated behavior and the rest of the world. An explainer from @hwchase17
-
Google accused of lazy benchmaxxing of models
By
–
Nah, thankfully Google is lazy and just benchmaxxing their models at this point
-
Evaluating Memory in Long-Horizon AI Agent Systems
By
–
LongMINT Evaluating Memory under Multi-Target Interference in Long-Horizon Agent Systems
-
User still rants about model merging in 2026
By
–
Still ranting about model merging in 2026 ✅ https://t.co/vGtNr0KpSc
— Maxime Labonne (@maximelabonne) 21 mai 2026Still ranting about model merging in 2026
-
Controlling LoRA Strength and Combining LoRAs in Image Tools
By
–
once the LoRA is trained, you can use it on the Image tool.
— Krea (@krea_ai) 21 mai 2026
you can control its strength and combine multiple LoRAs together. pic.twitter.com/90PJ6ONvYdonce the LoRA is trained, you can use it on the Image tool. you can control its strength and combine multiple LoRAs together.