Anthropic upgraded Claude Fable 5.2 and Mythos 5.2 with stronger coding, reasoning, and vision, plus faster output at lower prices. Fable is available broadly with cybersecurity and biology safeguards, while eligible organizations can apply to use Mythos for cybersecurity or biomedical research. (anthropic.com)
AI-generated code should be open source, the author argues. Open access would let others inspect, reuse, and improve the code. (jakub.jankiewicz.org)
The daily digest
Today's best Hacker News stories, summarized and screenshotted, one email a day.
A Google software engineer used 48 AI agents to scan about 1,200 km² of NASA lidar data and flag a previously unreported Maya site in Mexico. The search took less than 72 hours and identified a pyramid, plaza, and platforms beneath dense forest. (pietromiotti.com)
The daily digest
Today's best Hacker News stories, summarized and screenshotted, one email a day.
United Airlines is blocking AI agents from booking flights for travelers, citing risks to personal data and unintended purchases or account changes. The airline also warned that agents handling tasks such as check-in could conflict with federal security rules. (cbsnews.com)
The daily digest
Today's best Hacker News stories, summarized and screenshotted, one email a day.
Jonathan Blow asks why software has not changed despite claims that AI can solve coding. The provided content contains no further details about his argument. (youtube.com)
Apple is developing Harbor, a Mac app that runs open-weight AI models locally, keeping data on the user’s hardware rather than sending it to the cloud. It can connect multiple Apple silicon Macs over Thunderbolt 5 to pool memory and handle larger models. (9to5mac.com)
Anthropic says Claude Fable 5.2 and Mythos 5.2 improve coding, reasoning, and vision, with stronger performance on complex technical work and outputs up to 30% faster. Fable is available to all users with cyber and biology safeguards, while Mythos access is limited to approved research organizations. (anthropic.com)
Qwen4-Max is a new model in the Qwen family, with a focus on balancing efficiency and quality. Flash and 27B versions are also planned. (qwen.ai)
Amazon replaced Sonnet 4.6 with Strands Decider and says the change saves millions. The available content provides no further details on the savings or implementation. (strandsagents.com)
Fable 5.2 built a playable, multi-level beat ’em up from a child’s drawings and game ideas. The author says the game was fun and matched the nephew’s vision. (notesbylex.com)
Gemini 4 Hydrogen is a model in Google’s Gemini AI lineup. Google’s page introduces the model. (blog.google)
Claude is investigating authentication problems causing elevated errors across Claude.ai, Claude Code, Claude Console, and Claude Cowork. The incident was reported on November 13, 2026, and updates are pending. (status.claude.com)
AlphaProof uses AI and formal code to solve mathematical problems through proof search. The provided page text contains no further details about its capabilities or results. (deepmind.google)
Google’s Gemini Decisions API is a decision-making model and API inspired by Jev. It accepts audio and video inputs, offering capabilities beyond text-based decision models. (blog.google)
Google Antigravity is adding the Gemini 4 Hydrogen model. Its new Orchestrator option combines Hydrogen with Gemini 4 Carbon to balance quality, efficiency, and speed. (antigravity.google)
Smolmaxxing explores ways to run different LLMs on small, older devices. Its repository includes experiments, benchmarks, and device-specific tools. (github.com)
Anthropic’s Claude Fable 5.2 scored 65 on Artificial Analysis’s Intelligence Index, ranking first among 74 models and well above the 40-point median. It supports text and image input with a 1-million-token context window, but evaluation costs $26.54 per task, making it expensive compared with peers. (artificialanalysis.ai)
Anthropic CEO Dario Amodei’s prediction that AI could help cure cancer, Alzheimer’s, heart disease, and mental illness within a decade drew sharp criticism from experts and Eli Lilly’s CEO. Gary Marcus calls the claims unrealistic and compares them to Elizabeth Holmes’s medical promises. (garymarcus.substack.com)
AI-generated misinformation and fake books are making it harder for readers to tell what’s real and causing authors to answer for things they never wrote. The resulting confusion threatens the trust that connects writers and readers. (theatlantic.com)
E-means is a modern k-means algorithm that reaches lower loss in far fewer training epochs than Lloyd’s method. Benchmarks report 10-45× faster GPU runtime and 8-15× faster CPU runtime, and the code is available in Faiss. (e-means.github.io)
Analysis of six models from DeepSeek, Moonshot AI, MiniMax and Ant Group found weight patterns matching NVIDIA Blackwell GPUs, while earlier models matched Hopper chips. The findings support access to Blackwell hardware but do not establish where it was used or whether export controls were violated. (unemploymentlabs.com)
Claude launched Whiteboards and Sheets in beta, letting teams brainstorm together on shared boards and build or edit spreadsheets with Claude. Claude Dashboards is now Claude Data, with data stories, certified queries, and live sharing; all three are available on paid plans. (claude.com)
Jane Street is trialing new software engineering interview rounds for campus candidates, with less code-writing and more focus on design, code quality, and using LLMs. Most interviews will remain traditional coding sessions without LLMs, and candidates will be told in advance if they have a new-format round. (blog.janestreet.com)
Simon Willison used Claude Code to build Beep Locator, a web app that uses multiple phones and computers as microphones to estimate where an intermittent beep is coming from. He tested and refined the system with calibration sounds and hand claps, then used it to narrow the beep’s location without unpacking the room. (simonwillison.net)
AI agents now handle much of coding, debugging, writing, and even architecture, but engineers still shape the results. Their advantage comes from defining the right problem, judging complexity, and considering whether solutions will hold up over time. (twitter.com)
AI coding tools are rapidly changing software development, challenging the old idea that computers must follow rigid, explicit rules. The essay reflects on this shift through Emacs and the history of programming workflows. (commandlinesystems.com)
Meta introduced Muse Ember, a compact model built for reasoning, coding subtasks, tool use, and other repeated agent tasks. Meta says it matches Claude Haiku 5.5’s top score at roughly one-third the cost and outperforms every tested GPT-6 Luna setting. (research.meta.ai)
Google announced that Gemini 4 Carbon is now available. The announcement provides no details here about the model’s features or access. (blog.google)
Edge0 has open-sourced its 145B and 120B AI models, plus tools for running them locally. The company reports that Edge0-145B uses 29GB of peak memory and can handle coding, reasoning, and agent tasks on compatible laptops, including offline. (edge0.ai)
Data centers drove $8.25 billion in power infrastructure costs in 2025 but paid only 38%, leaving residents to cover most of the bill. Lower-income households, especially in rural areas, face the heaviest burden, with added energy costs reaching 28% of income for the poorest households. (news.cornell.edu)
AI-assisted app creation has flooded Apple’s review queue, delaying music apps and plugins for weeks instead of the usual one or two days. Developers including Pittsburgh Modular and Igor Vasiliev say the backlog has disrupted planned releases and cross-platform launches. (synthanatomy.com)
Brave Search API now supports pay-per-request payments through the x402 protocol, letting developers and AI agents search without API keys or accounts. Payments use USDC on Base, with other networks and stablecoins planned. (brave.com)