The 12 AI stories that mattered this week
Ranked by impact from 259 stories AIPulsee tracked so far this week. Each one explained in plain English.
AI unlocks Gödel's notebooks, said to contain hundreds of proved theorems
Finnish logician Jan von Plato led a Helsinki University project to decode 1,009 pages of Kurt Gödel’s 1940s notebooks written in the obsolete German Gabelsberger shorthand. After slow manual progress, he used the AI model Claude, which within days produced a comprehensive digest showing Gödel had proved hundreds of previously unknown theorems and anticipated results from the 1970s‑80s.
Gemini 4 Hydrogen and Orchestrator
Gemini 4 Hydrogen and Carbon models are combined in Google Antigravity to improve coding and agentic workflows. Gemini 4 Hydrogen is now available for paid subscribers.
Gemini 4 Hydrogen
Google announced Gemini 4 Hydrogen, a fast, cost‑efficient AI model aimed at everyday coding and agentic workflows. It is priced at $1.50 per million input tokens and $7.50 per million output tokens across a 1 million‑token context window.
I'm making LLMs work on small and old devices
A developer has created a project called smolmaxxing to run large language models on small, old devices like TV boxes and routers. The system uses custom kernels and memory techniques to keep only part of a model in RAM while streaming the rest from flash storage.
United Airlines bans AI agents from booking flights on behalf of travelers
United Airlines has banned artificial intelligence agents from making flight reservations on behalf of travelers, citing security risks and potential violations of federal rules. The airline says that AI agents would need access to customers’ personal and payment information, which could expose sensitive data and lead to unintended purchases or mileage use.
My AI agents found an unmapped Maya site?
A developer used multiple AI agents to analyze public NASA lidar data from the Maya region overnight. The agents identified a hilltop site with straight edges, a plaza, and a pyramid that was not in Mexico's official INAH register.
Gemini 4 Carbon is now available
Google announced that Gemini 4 Carbon is now available to developers, enterprises, scientists, and consumers via the API in Google AI Studio and Gemini for Business, priced the same as Gemini 4 Argon. Carbon is described as the strongest model with frontier performance in cybersecurity, complex coding, debugging, and scientific research, and it ranks #1 on several independent benchmarks while offering lower cost per task.
Claude Fable 5.2 scores 65 on Artificial Analysis, highest so far
Claude Fable 5.2, the latest version of Anthropic’s Claude series, achieved a score of 65 on the Artificial Analysis benchmark, the highest result recorded to date for any model on that test. The score reflects significant improvements in reasoning, problem‑solving, and language understanding compared with previous releases.
Claude Fable 5.2 and Mythos 5.2
Anthropic announced the release of Claude Fable 5.2 and Mythos 5.2, the latest upgrades to its flagship conversational model (Claude Fable) and its specialized research model (Mythos). The new versions deliver stronger reasoning, higher factual accuracy, improved safety mitigations, and faster, cheaper inference compared with the previous 5.1 releases.
Google Gemini routes personal Drive files and emails to human annotators
Google’s Gemini AI model is set up to automatically forward users’ private Google Drive documents and email contents to human reviewers for labeling and quality checks. This means that personal files and messages could be read by people employed by Google or its contractors without the users’ explicit consent.
Show HN: New Cartesia Text-to-Speech Model
New AI development in today's tech news — click source for full details.
Chicago schools to Google – we'd rather lose Search than be forced to accept AI
Chicago Public Schools (CPS) has decided to avoid using Google’s AI services by opting for a different search solution, after Google threatened to remove its search engine if schools did not accept AI. Google’s education VP Chris Phillips said schools would lose search access if they refused AI, citing deep integration of Search and AI.