ai model on Hacker News

107 titles from 93 authors in the last 30 days, with 102 distinct links and 13 Show HN.

Counts titles, it does not confirm facts. Open the source before concluding anything.

Projects on this term

The score comes from our model.

• Show HN: Lumabri – Run large AI models across your computers and servers.Pure C · score 7.3

  Lumabri is a C-based distributed inference engine that runs large AI models across multiple computers on a LAN, using a layer-wise sharding approach with explicit memory budgets and donor approval. It provides a terminal UI and bundled web chat for managing and using the swarm, with careful attention to security and resource accounting.

• Show HN: GhostBench – do AI models recommend dead SaaS? I tested my own thesis · score 7.3

  GhostBench/kyv-attest defines a signed, machine-verifiable attestation convention for vendor aliveness and operational facts, anchored in a transparency log, to help AI agents determine whether a SaaS vendor is still real. The accompanying benchmark shows that dead SaaS domains often return HTTP 200 and that small models misjudge vendor viability, motivating signed continuity as a stronger signal. It is interesting because it targets a real gap in agent infrastructure—vendor trust and liveness—rather than another API wrapper.

• Show HN: NullPrime – do AI models know when they can't know? · score 7.0

  NullPrime is a benchmark dataset and protocol for testing whether AI models can recognize when a problem is unlearnable from its representation. It constructs prime/composite classification tasks where all small-prime divisibility signals are filtered out, making the label deterministic but unlearnable from features, and uses this as a negative control for LLM calibration and abstention. The project includes a pre-registered LLM study showing most models confidently answer at chance on these hard numbers.

• Show HN: Parsimony – an open-source benchmark for how much code AI models add · score 7.0

  Parsimony is an open-source benchmark that measures how much code AI coding models add by parsing and counting syntax units in patches from public benchmarks like SWE-bench. It provides a reproducible leaderboard that normalizes for task selection, offering a novel perspective on model efficiency beyond solve rate. The project is well-documented and technically non-trivial due to multi-language analyzers and fair ranking methodology.

Recent titles

1. Show HN: NullPrime – do AI models know when they can't know?
   2026-10-31 · github.com · HN discussion
2. Open-weight AI models can exploit the patch gap
   2026-10-31 · epoch.ai · HN discussion
3. Rival AI models write an encyclopedia and judge each other blind
   2026-10-30 · scio.md · HN discussion
4. We asked eleven AI models to chair the drafting of a multilateral statement
   2026-10-29 · karlsnotes.com · HN discussion
5. An economy for open weight AI models
   2026-10-29 · twitter.com · HN discussion
6. OpenAI no longer training AI models on live internet, citing 'significant risks'
   2026-10-28 · techcrunch.com · HN discussion
7. Access Multiple AI Models in Your Python Apps Using OpenRouter
   2026-10-28 · kdnuggets.com · HN discussion
8. Giotto Sovereign European AI Models
   2026-10-27 · giotto.ai · HN discussion
9. At 14 Justine Tunney published software. Now she makes AI models run from a file
   2026-10-27 · timesofindia.indiatimes.com · HN discussion
10. Nvidia backs Indian startup helping companies run AI models with fewer GPUs
    2026-10-27 · techcrunch.com · HN discussion
11. I tested 12 small AI models on 2,500 design-system questions
    2026-10-27 · designsystemscollective.com · HN discussion
12. What to do if you discover AI models trained using your copywrighted book?
    2026-10-27 · HN discussion
13. Do you want to keep your job? Stop feeding AI models
    2026-10-27 · HN discussion
14. AI Models Can Provably Hide Arbitrary Capabilities
    2026-10-26 · maccus.fr · HN discussion
15. US political candidates tailor messages for AI models used by voters
    2026-10-26 · reuters.com · HN discussion
16. Cross-request KV cache exposure in the GLM-5.3 Workers AI model (155 responses)
    2026-10-26 · blog.cloudflare.com · HN discussion
17. AI models cost more when you don't use English
    2026-10-26 · restofworld.org · HN discussion
18. Show HN: Websites designed by 10 AI models, ranked by 4,500 human votes
    2026-10-26 · humansignal.com · HN discussion
19. Concerning behavior from a popular open-weight AI model
    2026-10-26 · cohere.com · HN discussion
20. Show HN: Mantid, a Mac app that lets non-ML people train their own AI models
    2026-10-24 · mantid.dev · HN discussion
21. Why Do AI Models Tell Left-Wing Voters to Support the Communist Party?
    2026-10-23 · fsi.stanford.edu · HN discussion
22. AI models think humans are a bunch of amoral sheep
    2026-10-23 · theregister.com · HN discussion
23. Cerebras Deploys Trillion-Parameter AI Models
    2026-10-22 · cerebras.ai · HN discussion
24. Every AI model has the same "controversial" take on humans
    2026-10-22 · modelcompare.ai · HN discussion
25. Show HN: Lumabri – Run large AI models across your computers and servers.Pure C
    2026-10-22 · github.com · HN discussion
26. France's OVHcloud plans frontier AI models to become Europe's second LLM player
    2026-10-21 · reuters.com · HN discussion
27. Alan Turing Institute wants to build a UK AI model from scratch and make it free
    2026-10-21 · hotminute.co.uk · HN discussion
28. The Debate over AI Model Welfare
    2026-10-21 · spectrum.ieee.org · HN discussion
29. Think App store shared revenue but for AI models
    2026-10-20 · dismodels.com · HN discussion
30. Testing Jev, a fast multiple-choice AI model, on forensic accounting tasks
    2026-10-20 · case-trail.com · HN discussion

All trends