OpenAI announced that an internal AI model, still under development, reportedly solved over 100 long-standing open mathematical problems in just 24 days. This includes a claimed solution to the existence and regularity of the Navier-Stokes equations, one of the Clay Mathematics Institute's Millennium Problems. The AI formalized its proof using Lean, a formal verification system.

This extraordinary claim sparked significant debate within the mathematical community. Notably, 25 Fields Medal winners issued a statement titled "A Severe Misalignment of AI in Mathematics," criticizing the use of open problems as mere benchmarks and highlighting the lack of new ideas generated during the process, proper attribution, and the training of future researchers.

In response, OpenAI established an "Advisory Group on Mathematics and Artificial Intelligence" at the Institute for Advanced Study in Princeton. This group, comprising nine prominent mathematicians, will advise on reviewing, valuing, and communicating new results, and recommending academic standards. While independent and without veto power over OpenAI's research pace, it aims to bridge the gap between AI labs and the mathematical community.

👉 Read the full article on Gizmodo

Google has integrated its latest AI model, Gemini Omni 1.1 Flash, into Google Vids, enabling anyone with a Google Account to create high-quality HD videos for free. This tool empowers users to generate compelling videos for business, community initiatives, or personal projects without requiring professional editing skills or a large budget.

Gemini Omni 1.1 is designed to give users precise control over their video narratives. It allows for extending scene durations smoothly while maintaining visual consistency, setting specific clip durations, and generating new AI video scenes in full 1080p HD or upscaling existing AI clips for seamless integration across the timeline.

To ensure transparency, every clip generated with Omni 1.1 in Google Vids includes an imperceptible SynthID digital watermark, allowing viewers to verify that the content was AI-generated. Additionally, upcoming features include Gemini 3.8 Flash-Lite text-to-speech, which will convert any script into natural-sounding voiceovers across more than 100 languages.

👉 Read the full article on Google Blog

Anthropic announced that its AI model, Claude, has discovered a novel enzyme system featuring CRISPR-like repeats. This discovery represents an early success for Anthropic's newly formed life sciences research group and laboratory, which aims to establish a new paradigm for biological research where AI agents collaborate with humans to accelerate discoveries.

In this research, approximately 950 Claude agents spent 21 hours sifting through a massive database of DNA sequences. With only high-level direction from human scientists, Claude identified an unusual pattern: a tandem repeat array adjacent to a reverse transcriptase (RT) gene. This system, named "array-associated reverse transcriptases (ART)," was found in bacteriophages.

The ART system comprises three parts: the RT, a partner gene, and a long array of evenly spaced DNA repeat sequences, with the repeat layout resembling a CRISPR array. This finding demonstrates AI's capability to autonomously detect anomalies and drive analyses to initiate biological discoveries, showcasing the value of AI-driven hypothesis generation to the broader scientific community.

👉 Read the full article on Anthropic