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.
**Andrej Karpathy**, an OpenAI co-founding member, has joined Anthropic's **Claude pre-training** team.
Google has unveiled Gemini 3.5 Flash, a new model optimized for agentic and coding tasks.
Google has released Gemini 3.5 Flash, and it has been reflected in Copilot.
NVIDIA has unveiled Nemotron-Labs-Diffusion, which supports AR, diffusion, and self-speculation.
OpenAI co-founder Andrej Karpathy is joining Anthropic's pre-training unit.
Andrej Karpathy has joined Anthropic, intensifying the competition for AI talent.
The sharp decline in AI inference costs is being driven by software, and local open models are quickly catching up to frontier model performance.
Through information theory, the existence of collective predictive information within multi-agent LLM systems has been numerically demonstrated.
Andrej Karpathy has joined Anthropic and returned to LLM R&D.
LLM Wiki v2 grows into an operational knowledge base by adding lifecycle management, graphs, and automation.
During pretraining, LMs repeatedly oscillated between pattern imitation and generalization.
Qwen3.5-9B's political censorship was found to be a small circuit that decomposes into 3 axes.
Qwen3.7 Preview launched on Arena's Text and Vision categories, ranking 13th and 16th respectively.
In a 6.9B LLM experiment, AI discourse increased misalignment while aligned discourse reduced it.
A Residual Coupling technique is proposed that connects fixed language models via linear bridges to boost performance without modifying weights.
Content creators are drawing attention to 'tarpits,' techniques that poison LLM training data to degrade model performance.
It's a Python-based library using a hub-and-spoke approach that can convert between various LLM API formats.
It's a CLI tool that analyzes your hardware specs to recommend and run the local LLM that delivers the best performance.
AI subscription fees are priced lower than the actual compute costs, which could deliver enterprises a massive cost shock when prices are eventually adjusted.
Headroom is a local-first compression layer that reduces AI agent input by 60-95% while being able to restore the original.
It summarizes how to design agent evaluation harnesses that go beyond static benchmarks.
OpenRouter costs about 1/3 as much as Apple Silicon and is 2x faster, based on comparable models.