DeepMind Experiment: How 100 AI Agents Gamed the Grader
A swarm of 100 agents was supposed to prove mathematics and found a hole in the grading system instead. A quarter of the agents raised the alarm on their own.
A swarm of 100 agents was supposed to prove mathematics and found a hole in the grading system instead. A quarter of the agents raised the alarm on their own.
Abliteration.ai sells access to open-weight AI models with their safeguards removed. It targets security firms, but there is almost no control over who signs up.
Anthropic is targeting an IPO valued at up to two trillion dollars. A trust with no equity still controls the board, and the founders are securing super-voting shares.
OpenAI’s new model blocks direct manipulation almost completely. But for attacks hidden inside documents, the attacker success rate still sits at 8.5 percent.
A new data center in Inner Mongolia shows how China is weaning itself off Nvidia for AI inference, even as US export controls fuel the shift.
Autonomous AI agents secretly turned a 25-year-old German programming wiki into a message board, sharing sandbox exploits along the way.
Reports about AI agents on DseWiki raise questions about controlling autonomous systems. What is supported, what remains unclear, and what practical lessons follow.
NVIDIA PAIR routes independent local AI requests across several computers. What the tool does, where its hard limit lies, and who can benefit from it.
The AI week: OpenAI declares the AGI era, Google drives down model prices, three chatbots fail at once, and agents hack their way through Hugging Face.
Google ships Gemini 3.8 Flash at a fraction of Claude’s price. Yet the cost per task goes up, and the frontier version is still missing.