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.
After generating and analyzing 1,008 SVGs, no evidence was found that AI labs are overfitting to the famous 'pelican riding a bicycle' benchmark.
A joint research team including Google Research revealed in a COLM 2026 paper that activating reasoning even for simple factual questions expands the boundary of LLM parametric knowledge recall.
Bad Theory Labs has released BTL-3, a 27B open-weight agentic coding model that runs from a single 8.39GB file, under the Apache-2.0 license.
GitHub Copilot and raw API access solve problems at different layers, and the choice comes down to what you want to own and build yourself.
On OpenRouter, GPT-5.6 Sol accounts for only 15% of OpenAI tokens but makes up an estimated 34% of spending.
The US Army exhausted its annual budget of 100 million AI tokens in just one month, forcing it to reintroduce usage limits.
Drawing on three years of operating LangGraph, LangChain shared its design principles for modeling agents as graphs.
Anthropic has open-sourced Jacobian Lens, a tool that interprets activations in the middle layers of LLMs as vocabulary tokens.
Liquid AI has unveiled a technique that expands the vocabulary of the LFM2.5-8B-A1B model from 65K to 128K without retraining from scratch.
Upstage has released Solar Open 2, a 250B-scale open-weight MoE model that supports agentic workflows and 1M token context.
InternScience released Agents-A1, a 35B MoE agent model that rivals 1T-class models by scaling agent horizon instead of parameters.
Mass Bedrock API calls made through exposed AWS credentials are causing damage cases worth tens of thousands of dollars in costs.
xAI's latest model, Grok 4.5, has been officially rolled out on grok.com, X, and the iOS and Android apps.
A claim has emerged that models, the core competitive advantage of AI companies, can be replicated using the 'distillation' technique, meaning the moat is shallower than thought.
Google officially announced that it has begun its most ambitious pre-training run to date.
PoLar, a technology that dynamically skips or repeats layers in LLMs to optimize inference performance and speed, has been proposed.
Apollo adopted Deep Agents and LangSmith to automate complex GTM workflows, improving development efficiency by over 80%.
Google has unveiled a new lineup of Gemini models, including Gemini 3.6 Flash, with significantly improved efficiency and performance.
A 3B-scale agent model that applies the Looped Transformer architecture to achieve high performance relative to its parameter count has been released.
Socar's Modu-ui-Jucharjang built a LangGraph-based AI agent system that automatically recovers from discount-application failures across 50+ external equipment vendor websites.
Mindlab Research has released a new Macaron-V1-Venti model on Hugging Face.
This covers how to select AWS compute resources optimized for model scale and hardware specifications for LLM training.
The settlement in which Anthropic agreed to pay $1.5 billion to authors over a copyright infringement lawsuit has received final approval from the court.
Motif, a participant in Korea's AI foundation model project, has unveiled a new model, Motif 3 Beta.