can't on Hacker News

102 titles from 97 authors in the last 30 days, with 97 distinct links and 14 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: UOS – a workspace the server can't read (post-quantum, E2EE) · score 8.3

  uOS is a browser-based end-to-end encrypted workspace suite (chat, mail, docs, video, etc.) that uses post-quantum cryptography (ML-KEM, ML-DSA) to ensure the server operator cannot read user content. It extends a unified encryption model across all applications and offers organisation features like key escrow and compliance controls without compromising operator blindness.

• 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: Cascache – A Go cache where slow reads can't undo invalidation · score 7.0

  Cascache is a Go read-through cache that prevents slow in-flight loads from resurrecting stale values after invalidation, using a per-key fence token (similar to a generation counter). It supports local, Redis, and shared-process backends, multiple value stores and codecs, and typed keys with randomized TTLs. The fence-based invalidation approach is a coherent, well-engineered fix to a real cache race condition.

• Show HN: Pith – a proxy that cuts LLM output tokens, and refuses when it can't · score 7.0

  Pith is a proxy that reduces LLM output token costs by automatically testing cheaper output profiles and only pinning them when a judge confirms they match baseline quality. It fingerprints routes, records samples, replays under candidate profiles, and refuses to change anything if the savings don't justify the sweep cost or if quality degrades. The refusal-first design and multiple integration points (standalone proxy, LiteLLM, Headroom, Portkey) make it a practical cost-optimization tool for API-based LLM applications.

Recent titles

1. Sell a developer problem that LLM's can't solve?
   2026-11-01 · HN discussion
2. You can't create a different life while continuing to make the same choices
   2026-11-01 · randomfuckingknowledge.substack.com · HN discussion
3. Stuck on an island for years, can't get transit visa to fly home
   2026-11-01 · bbc.co.uk · HN discussion
4. Show HN: NullPrime – do AI models know when they can't know?
   2026-10-31 · github.com · HN discussion
5. Show HN: Cerebellum – small model on the robot, big model for what it can't undo
   2026-10-30 · github.com · HN discussion
6. British intelligence misplaced 1000s of storage devices still can't say how many
   2026-10-30 · theregister.com · HN discussion
7. Show HN: Send21 – Crypto pay links an AI agent can create but can't spend
   2026-10-30 · send21.io · HN discussion
8. Show HN: Korea-data-MCP – Korean public APIs that agents can't misread
   2026-10-30 · github.com · HN discussion
9. The Web Can't Afford a JavaScript Runtime Monoculture
   2026-10-30 · redmonk.com · HN discussion
10. Mexico is a showroom for electric vehicles you can't buy in the U.S.
    2026-10-29 · washingtonpost.com · HN discussion
11. My Students Can't Read
    2026-10-29 · chronicleofhighereducation.substack.com · HN discussion
12. You Can't Automate Caring
    2026-10-29 · ericfriese.substack.com · HN discussion
13. The work we can't delegate
    2026-10-29 · mozestudio.com · HN discussion
14. Show HN: MealWords – Show waiters what you can't eat, in their language
    2026-10-29 · mealwords.com · HN discussion
15. Total Recall: The Woman Who Can't Forget
    2026-10-29 · wired.com · HN discussion
16. China's top-secret spaceplane can't hide in low-Earth orbit
    2026-10-28 · arstechnica.com · HN discussion
17. Show HN: Wenlan – your LLM wiki that flags what it can't back up
    2026-10-27 · github.com · HN discussion
18. Marissa Mayer Can't Keep Pretending 'Gender Doesn't Matter'
    2026-10-27 · 404media.co · HN discussion
19. Static reasoning can't prove runtime behavior: benchmark of AI PR reviewers
    2026-10-27 · heal.dev · HN discussion
20. A Salesforce AI agent that can't publish a hallucination
    2026-10-27 · github.com · HN discussion
21. What a Wayland compositor can and can't protect you from
    2026-10-26 · demiguru.codeberg.page · HN discussion
22. A Man from the Future: Why Johnny Can't Stop Abstracting
    2026-10-26 · muratbuffalo.blogspot.com · HN discussion
23. How to Build a Reality You Can't Escape – Walt Disney [video]
    2026-10-26 · youtube.com · HN discussion
24. Why it is foolish to think AI can't write like humans
    2026-10-26 · HN discussion
25. Why Anthropic can't afford to delay its IPO
    2026-10-26 · axios.com · HN discussion
26. Altman issues self-serving warning they've created a monster they can't control
    2026-10-24 · the-decoder.com · HN discussion
27. Show HN: A phone lock I can't undo on impulse because I never learn the code
    2026-10-24 · tryscreenfasting.com · HN discussion
28. I can't believe we're still arguing over whether X is for jerks
    2026-10-24 · lexontech.org · HN discussion
29. In the biotech world, you can't see everything
    2026-10-24 · technologyreview.com · HN discussion
30. The animals you can't outrun – and the ones you can
    2026-10-24 · bbc.com · HN discussion

All trends