Keep pulling the thread on United States.
The latest frontier AI models have achieved parity with expert human professionals on legal tasks as measured by the GDPVal benchmark.
Anthropic recently retracted its previous safety commitments, which had stated the company would pause development if certain capabilities could not be developed safely.
Anthropic is in open conflict with the U.S. federal government over the potential military application of its AI models.
Google's Gemini family of models has a 1 million token context window.
U.S. chip export controls are limiting the resources of Chinese AI companies, causing them to fall behind American competitors in developing frontier models.
Researchers at Anthropic were able to identify and manipulate the internal representation of the "Golden Gate Bridge" concept within a Claude model, causing it to obsessively discuss the landmark.
A virtual lab led by an AI agent, with AI coworkers it created, successfully designed new nanobodies to treat emerging variants of the COVID-19 virus.
On a benchmark of real-world freelance tasks from Upwork, the percentage of total contract value that AI agents could successfully complete increased from 8% with GPT-4 to over 80% with the latest models in about 18 months.
Three different AI models won gold medals at the International Mathematical Olympiad (IMO).
Mathematician Terence Tao has reported that AIs are now solving previously unsolved Erdos problems.
A Google AI model discovered a new immunotherapy approach for cancer treatment.
OpenAI CEO Sam Altman predicts the company will have an intern-level AI researcher by 2026.