Notes by Jin

Models & Research Web Content Pollution Poses a Threat to AI-Powered Recommendation Systems — A new benchmark called FORGE tests how vulnerable search-augmented LLMs are to fake product promotion when retrieval results are polluted, finding that even a single compromised page produces fooled rates of up to 27% across 12 models,

Policy & Society Anthropic Proposes Government Framework for Catastrophic AI Risk — Anthropic has released a policy proposal urging governments to take legal authority to block or prevent deployments of the most powerful AI models that could pose catastrophic risks. The framework outlines how regulators should respond as AI capabilities accelerate

Models & Research Decart Launches Oasis 3 World Model for Photorealistic Autonomous Driving Simulation — Decart has released a real-time world model capable of generating extended photorealistic driving environments, made available through an API for autonomous vehicle developers. The system aims to enable more efficient and scalable testing for self-driving applications.

Industry & Funding OpenAI Files Confidential IPO Registration with SEC — OpenAI has submitted a confidential S-1 filing to the SEC, following a similar move by rival Anthropic just weeks earlier. The company has not yet announced a timeline for proceeding with a public offering. OpenAI ↗ My Takeaway: As two leading

Industry & Funding AI Token Prices Set to Rise as Companies Eye IPOs — Major AI firms are expected to increase the cost of AI usage, with price hikes likely tied to preparations for going public. The trend signals a shift in how AI companies plan to monetize their services as

Models & Research Google's May 2026 Recap: The Agentic Gemini Era Arrives — May was Google's biggest AI month of the year, headlined by I/O 2026 and a clear theme: agents that act, not just answer. Two new models lead — Gemini 3.5 for executing multi-step

Industry & Funding Google Commits Nearly $1B Monthly to SpaceX for Compute Power — Google has agreed to pay SpaceX roughly $920 million each month for computing resources, a deal driven by unexpectedly high demand for its recently launched AI offerings. The arrangement underscores the enormous infrastructure investments now required to

Models & Research AI Chatbots May Be Eroding Cognitive Self-Regulation — A psychologist who has spent decades studying human-technology interaction raises concerns that reliance on AI chatbots could diminish people's ability to manage their own thinking and attention. The research highlights growing questions about the long-term neurological and behavioral

Models & Research Audio-Language Models Override Audio Evidence When Conflicting Text Is Present — Researchers investigated whether audio-language models suppress audio-supported answers or simply fail to represent them when conflicting text is provided, finding that correct audio-grounded responses are often available but overridden rather than absent. The study uses counterfactual probing