Hacker News Digest

· 12 Hacker News

  1. ▲ 132Tim Schafer's 1996 Grim Fandango puzzle doc hides a joke box for your tears of joy

    The original puzzle design spec for Grim Fandango, posted by Tim Schafer at Double Fine for the game's tenth anniversary, has resurfaced on Hacker News. Commenters trade memories of the LucasArts cla…

  2. ▲ 83Amiga Unix returns: community project brings Amix to 68040/68060 and modern accelerators

    Amiga Unix (Amix), Commodore's System V Release 4 from 1990–92, is being revived by amigaux.org. The community project supports 68040/68060 machines and modern accelerators, with a modern toolchain,…

  3. ▲ 83Visa Lottery Study: Cutting Low-Skill Immigration Shrinks Firms Without Adding American Jobs

    A randomized H-2B visa lottery reveals that restricting low-skill foreign workers makes firms smaller, not more American jobs. Losing the lottery halved H-2B employment, cut revenue, and reduced inve…

  4. ▲ 82AI chatbots get financial answers wrong 'most of the time'

    A new study finds that leading AI chatbots, including those from OpenAI and Google, give incorrect answers to financial queries most of the time. The errors range from simple arithmetic to complex in…

  5. ▲ 44Mini-AGI trains a continual learning model from scratch on an 8GB laptop GPU

    Mini-AGI is a byte-level language model that assembles its own architecture, trains on a single 8GB VRAM GPU, and keeps learning without catastrophic forgetting. It pages expert weights from disk, gr…

  6. ▲ 42Every messy codebase hides deterministic cores you can defragment

    Gary Bernhardt's Functional Core, Imperative Shell is a classic, but pure functional programming isn't always practical. The author argues we can relax the requirement from purely functional to merel…

  7. ▲ 37Why Backpropagation Must Go Backward

    Backpropagation is often explained as propagating errors backward, but why not compute gradients forward? Gregory Gundersen shows that a forward pass would repeatedly propagate the same terms, leadin…

  8. ▲ 35Kev: Tiny decision models you can train and run on your own

    Kev is a family of small decision models (0.8B, 4B, 9B) built on Qwen3.5 that answer yes/no, multiple-choice, and rating questions in a single request. They run on CUDA and Apple Silicon, and you can…

  9. ▲ 14AI now writes 17.25% of all Linux kernel patches

    The Lunduke Journal reports that AI-generated code accounted for 17.25% of all Linux kernel patches in September, with 1,634 AI-written submissions in a single week. Commenters like DHH predict the s…

  10. ▲ 13ByteDance and Tsinghua's DAPO sets new bar for open-source LLM reinforcement learning

    DAPO, a fully open-source RL system from ByteDance Seed and Tsinghua AIR, achieves 50 points on AIME 2024 with a Qwen2.5-32B base model, surpassing DeepSeek-R1-Zero-Qwen-32B using only half the train…

  11. ▲ 12LLMs Are Obsessed With the Word 'Spine'—And It Shows in GitHub Pull Requests

    Hillel Wayne noticed AI-generated TLA+ specs kept using 'spine.' Digging into GitHub pull request titles, he found a 20x jump in 'spine' PRs in 2026 versus all of 2025—far outpacing overall PR growth…

  12. ▲ 8AI workers laugh off doomsday warnings: 'LLMs won't wipe out humanity because they just don't have that dog in them'

    Many AI employees at OpenAI, Meta, and DeepMind are skeptical of claims that unchecked AI development could kill everyone. They dismiss the warnings as vague and hypothetical, with reactions ranging…