Hacker News

New stories

Live mirror
30 storiesupdated just nowView source snapshot
  1. Asking Authors About Their Own Papers(medium.com/tmlrorg ↗)
    discuss
  2. Next-gen gravitational wave detectors could spot the first black holes(phys.org ↗)
    discuss
  3. US interest rates raised for first time in three years(bbc.com ↗)
    discuss
  4. My website charged AI agents a penny per page. I watched Claude pay it(suganthan.com ↗)
    discuss
  5. Show HN: Helping agents break out of their sandboxes in a monitorable way(sparrowsystems.co ↗)
    discuss
  6. Encouraging Deception in Compaction Summaries(alignment.openai.com ↗)
    discuss
  7. Jev-like mode for DiffusionGemma model(github.com/vllm-project ↗)
    discuss
  8. DeepSeek's Parent Company(high-flyer.cn ↗)
    discuss
  9. The Flask Mega-Tutorial, Part I: Hello, World(miguelgrinberg.com ↗)
    discuss
  10. Typesafe.ai Jev Open Source Alternative Qwen-2.5-1B-RLCD(huggingface.co ↗)
    discuss
  11. AI Debt vs. Treasuries(ft.com ↗)
    1comments
  12. "WeWorm": Zero-click worm could have taken over all WeChat accounts(heise.de ↗)
    discuss
  13. Dwarf Star Support for Qwen3.8 Flash Next(github.com/antirez ↗)
    discuss
  14. YuE2 – Frontier music generation with editable scores(github.com/multimodal-art-projection ↗)
    discuss
  15. Pactora – AI contract review built by a lawyer, not engineers(pactora.uk ↗)
    discuss
  16. Automattic says CEO Mullenweg was gone and back inside 33 hours. What happened?(thenewstack.io ↗)
    discuss
  17. OpenSpec – A lightweight and configurable AI spec framework(openspec.dev ↗)
    1comments
  18. Calculator for Australian take-home pay, tax and super(auspaycalculator.org ↗)
    discuss
  19. How much does one hour of bots cost on Cloudflare Workers?(gfuse.dev ↗)
    discuss
  20. Plans for Agents (and Humans)(hive.technology ↗)
    discuss
  21. Australia says it could follow Canada in forging deeper ties with EU(independent.co.uk ↗)
    2comments
  22. Fossil tooth analysis confirms Tyrannosaurus rex was warm-blooded(cnn.com ↗)
    discuss
  23. The mascots, design and visual history of Wikipedia(itsnicethat.com ↗)
    discuss
  24. 60%+ of Japan firms with foreign workers would struggle without overseas hiring(japantimes.co.jp ↗)
    discuss
  25. Show HN: DroidPutter – Run Cardputer Apps on Android via USB OTG(github.com/fcavalcantirj ↗)
    discuss
  26. Why building a Rust LSP is hard(rust-glancer.github.io ↗)
    discuss
  27. Some things Veloren does differently(jsbarretto.com ↗)
    discuss
  28. New Potential Automatic Parallelizer(deshi.ch ↗)
    1comments
  29. I Put My Startup Esdenera on Hold(archive.org ↗)
    discuss
  30. Am I Cited by AI? See which AI assistants are reading your papers(amicitedby.ai ↗)
    discuss

Assistant Benchmark

1 pointsby 1h agoassistantbenchmark.com
1 comments
54m agoHN ↗

I have been trying to build an assistant for a year+ at this point. I have still yet to come up with a model, harness, and set of instructions that get it to reliably not contact me just to regurgitate things I already know like text messages, emails, or calendar events. I want it to have those things so it can alert me to 2nd order situations.

For instance, lets say I have a meeting with Bob, but look Bob emailed me 30 minutes ago saying to bring XYZ to the meeting - that's worth notifying me. But pinging me to say "Nothing in the next 3 hours, good time to work on <overdue reminder from 3 months ago to take garbage out". Boggles my mind they can solve novel math problems but can't figure out what is worth my attention and what is not.