AI Briefing
Hacker News news
The latest AI and developer news from Hacker News, with short summaries.
Visit website →CS336: Language Modeling from Scratch
Stanford University has released its CS336 course, covering the entire process from designing Transformer models to training and evaluation.
NVIDIA Cosmos 3
NVIDIA has open-sourced Cosmos 3, a frontier foundation model for Physical AI.
Launch HN: Expanse (YC P26) – Optimizing Idle GPU Capacity
Expanse launches to reduce idle capacity by precisely predicting resource requirements for GPU clusters.
Hugging Face Scans 7.6 Petabytes of Training Data for Secrets
220,000 valid credentials were found in Hugging Face's public training data.
ChatGPT for Google Sheets Exposes Workbook Data Leak Vulnerability
A vulnerability enabling data exfiltration and phishing attacks via indirect prompt injection has been discovered in the ChatGPT for Google Sheets extension.
Odysseus – Self-Hosted AI Workspace
It is an open-source workspace that uses local models and APIs to build a personalized AI agent environment.
The Website Specification
A specification that defines the technical standards a website should meet for both AI agents and humans.
Ernst & Young Publishes Cybersecurity Report Full of Hallucinations
EY's cybersecurity report was found to contain numerous fake citations, statistics, and other LLM hallucinations.
OpenRouter Raises $113 Million Series B Funding
OpenRouter has raised a $113 million Series B round from investors including NVIDIA and Alphabet's CapitalG.
US companies begin limiting AI usage amid soaring costs
Companies are starting to limit AI usage in response to rising costs of adopting generative AI.
MCP is dead
A technical analysis has raised concerns that MCP excessively occupies context windows and has low operational reliability.
Robinhood Launches Support for AI Agent Stock Trading
Robinhood has launched a feature that lets AI agents directly trade and settle stocks.
Mistral AI Now Summit Paris Field Notes
Mistral AI declared its transition into a full-stack AI company spanning compute and platform, beyond just model development.
Liquid AI Unveils 8B-A1B MoE Model Trained on 38T Tokens
Liquid AI has released an on-device 8B-A1B MoE model trained on 38T tokens with enhanced reasoning capabilities.
Real-Time LLM Inference on Standard GPUs: 3,000 tokens/s per Request
This describes a technique that optimizes single-request decoding speed, a core metric for AI agents, achieving 3,000 tokens per second on standard GPUs.
AI Film Promoted as a "Cannes Film Festival Premiere" Worth $500,000 Was Not an Official Selection
It has been revealed that the AI film AI startup Higgsfield claimed was a Cannes Film Festival premiere was actually screened at a commercial market, not the official program.
Claude Code – Every Configuration Method Not in the Official Docs
An analysis of the Claude Code source code reveals advanced configuration and hook features not documented officially.
PickMysterious Hy3 LLM Dominates OpenRouter Model Rankings by a Wide Margin
Tencent's Hy3 model has surpassed Claude to take an overwhelming No. 1 spot on OpenRouter's usage rankings.
Biohub Unveils World Model for Protein Biology
Biohub has unveiled an AI-powered world model for predicting protein structures and designing new proteins.
Show HN: Ktx – An Open-Source Executable Context Layer for Data Agents
Ktx, an open-source context layer that helps data agents accurately query data warehouses, has been released.