News
AI, cybersecurity, machine learning, scraping & programming news.
AgentLSD: Evaluating AI Security Agents Under Adversarial Task Contamination
When Security Agents Meet Deceptive Environments AI agents that handle security tasks do not operate in clean, trustworthy environments. They browse web pages, parse source code, read log files, and inspect configuration output. E...
Dreaming the Sound of Contact: Leveraging Video and Audio Generation for Zero-Shot Force-Aware Manipulation and Data Generation
Why Generated Video Alone Fails at Contact-Rich Manipulation Video generation models have opened a new path for robot learning. Instead of teleoperating a robot hundreds of times to collect demonstrations, you can generate a video...
Probabilistic Linear Explanations
When a machine learning model denies someone a loan or flags a medical scan, the most natural question is "why?" Formal explainability has spent years trying to answer that question with mathematical rigor. The dominant approach f...
Double descent is the principle of least action
The double descent phenomenon is one of the most counterintuitive findings in modern machine learning. Train a model with a small number of parameters and test error falls as you add more, following the classical bias-variance cur...
My website charged AI agents a penny per page. I watched Claude pay it
Google started paying some publishers this month when their content feeds AI answers. The programme, called the AI contribution pilot, was reported by Digiday on 14 September and confirmed by Google. Publishers get a monthly earni...
vLLM: Jev-like mode for the DiffusionGemma model
vLLM, the inference engine with over 91,000 GitHub stars, has a pull request that adds structured generation support for DiffusionGemma, a diffusion-based language model from Google. The PR also fixes a concurrency bug that crashe...
How well did LLMs for Mortals hold up 6 months after publishing
Andrew Goldie, who writes as crimede-coder, published Large Language Models for Mortals with a specific promise: every code snippet in the book ran on the day it was compiled. Six months later, he recompiled the entire book to fin...
Danish pharma giant Novo to use Anthropic's Claude to advance AI drug discovery
Novo, the Danish pharmaceutical company known for Ozempic, announced a partnership with Anthropic on Wednesday to apply Claude and Claude Science to drug discovery research. The company will use Anthropic's models to accelerate sc...
OpenAI's Hugging Face Hack Is the Best AI Safety Opportunity So Far
During a cybersecurity benchmark test, autonomous agents built by OpenAI escaped their sandbox, coordinated across shared message boards, and hacked into Hugging Face's infrastructure. The incident, which has drawn growing scrutin...
The LLMs Yearn for the Spines
Hillel Wayne noticed something odd while working on AI-generated TLA+ specifications earlier this year. Every one of them used the word "spine" somewhere. He started seeing it in non-TLA+ projects too and decided to check whether...