AI Briefing#llm
The latest AI and developer news about #llm, with the original source and a short summary.
The latest AI and developer news about #llm, with the original source and a short summary.
METR's time horizon keeps growing, but we examine whether the hourly cost of AI agents is rising along with it.
The Swiss AI Initiative supports research on large-scale, public AI models.
MoDA improves depth scaling, boosting both performance and efficiency in a 1.5B model.
Speculative checkpointing has been merged into llama.cpp, and speed improvements have been confirmed in some tasks.
Claude Design shakes up Figma's canonicity, highlighting a code-centric workflow.
Unweight reduces LLM weights by 15-22%, freeing up about 3GB of VRAM on H100.
This piece examines the risks that rude prompts and sycophantic AI pose to collaboration and judgment.
Chatbots like ChatGPT and Gemini showed biased party recommendations in the context of a Danish election.
A calculator comparing why the same prompt becomes more expensive on Opus 4.7.
Lore 0.2.0 is released with ThinkingStream and non-destructive embedding migration added.
AI investment and performance have surged, but employment and public perception remain mixed.
A story of regaining LLM/Python instincts and codebase understanding through hand-coding
It compares the token usage difference between Opus 4.6 and 4.7 using anonymous inputs submitted by the community.
Qwen 3.6 35B-A3B has been quantized for Ollama, supporting even tools and thinking.
Boosted Qwen3.6-35B-A3B speed by 54% on an RTX 5070 Ti using --n-cpu-moe.
Z.AI's GLM-5.1 has been added to the NVIDIA API and is offered as a free endpoint.
A benchmark, safety, and tensor analysis comparing Qwen 3/3.5 with HauhauCS, Heretic, and Huihui was released.
On M3 Ultra, Qwen 3.6 35B achieved a 100% tool-calling success rate across most of the 5 agent frameworks.
On 28 Zod tasks, Opus 4.7 had the same pass rate as 4.6 but was faster and more efficient.
Unweight reduces LLM weights by 15~22% on H100 while keeping outputs bit-exact.
Qwen3.6-27B, an open-source sparse MoE model optimized for agentic coding, has been released.
mii-llm released the 0.4B multilingual models Zagreus/Nesso, trained on 64×A100.
Claude Opus 4.7 scored 74.0 on the structural biology benchmark, significantly ahead of 4.6.