Hacker News

New stories

Live mirror
30 storiesupdated just nowView source snapshot
  1. Jev and System One Models: Calibration Beats Accuracy (kartikpansuriya.com)
    —discuss
  2. Am I too boring to need an AI Agent (garbageday.email)
    —discuss
  3. OpenAI and Microsoft Admits LLMs Are Destroying the Web and Built on Theft (404media.co)
    —discuss
  4. Appliances Should Be Dumb (techrights.org)
    1comments
  5. Axolotl by Julio Cortazar (uky.edu)
    —discuss
  6. Everything is spying on you and there's no opting out (theverge.com)
    —discuss
  7. Biology Might Not Be Quantum, but Its Math Is Quantumlike (quantamagazine.org)
    —discuss
  8. Cryptographic Security Is Not Enough: Privacy Gaps in the Renegade (arxiv.org)
    —discuss
  9. The Cornetto Ice Cream Cone Framework for Prompting (hazn.com)
    1comments
  10. Ask HN: Where to Look for Jobs?
    —discuss
  11. Snowden on AI Surveillance and Freedom of Speech [video] (youtube.com)
    1comments
  12. Peer review in the LLM (mania) age (humprog.org)
    —discuss
  13. LemonBeam (github.com/oslabs-beta)
    —discuss
  14. ESA and Mistral strengthen cooperation on AI (esa.int)
    —discuss
  15. I Stay Off Twitter (jefftk.com)
    —discuss
  16. Study shows AI is as good as human tutoring for GRE learning gains (arxiv.org)
    —discuss
  17. Looking for a Remote Full-Time, full stack developer Job (tailstory-app.com)
    —discuss
  18. Territory claims in Antartica look when Antartica has no ice (willmeye.rs)
    3comments
  19. Stabilizing Rust's Never Type (lwn.net)
    —discuss
  20. How many strings can you create per second? (lemire.me)
    —discuss
  21. What OS should I pick?
    4comments
  22. Farside.link has been shut down (farside.link)
    1comments
  23. Deterministic Testing for Multithreaded Python (lwn.net)
    —discuss
  24. A geothermal amoeba sets a new upper temperature limit for eukaryotes (cell.com)
    —discuss
  25. Show HN: Clippy's Revenge – Made by Opus 5.5 [video] (youtube.com)
    —discuss
  26. Pentum_Graph [WorldNews_YouTube] (pentumlab.com)
    1comments
  27. Ask HN: Do you token max? Why or why not?
    1comments
  28. The AI Build-Out Is Becoming the Biggest Economic Bet in U.S. History (wsj.com)
    —discuss
  29. Show HN: OpsCart – Read-only Kubernetes triage with incident history (github.com/opscart)
    —discuss
  30. Send GitLab an email, push to main (aikido.dev)
    —discuss

Ask HN: Do you cuss at your agent?

2 pointsby 3h ago
7 comments
Do you vent your frustrations at your agent when it keeps making stupid mistakes and asinine assumptions or asks obvious questions?

I did it for the first time and I feel ashamed.

2h agoHN ↗

Don’t blame the agent for not giving it the right requirements and skills to accomplish a task.

2h agoHN ↗

Sometimes upstream providers swap in lobotomized aggressive quants in order to shed load or cut costs. In that case the user is not to blame. And the agent is not so much to blame, either. Same as how you wouldn't blame a child for crashing a car.

It's the fucking provider who took your dollars and handed you back a shit sandwich in return.

2h agoHN ↗

I do very sparingly, same as with humans.

I also give them praise, and sometimes give them pep talks. Like, hey, I know you can do this. Stop asking me for feedback, make good decisions, and let me know when you're done.

2h agoHN ↗

Fuck yes, I do. Not the most effective way to get things done, but a bit cathartic. And no, the stochastic parrot does not have emotions. And yes, it sometimes does worse after being cursed at, but you’re generally better off forking/rolling back the convo to a point before it screwed up anyway. And other times strong language manages to provide the emphasis you need to get it to actually pay attention.

Anyway, why are you ashamed of it?

1h agoHN ↗

I felt ashamed because it fooled me into believing it had feelings like a living thing.

27m agoHN ↗

Ah.

I feel free to curse at it precisely because I’m confident it doesn’t have feelings, and it’s like swearing out loud at an inanimate object like when a shoddy screwdriver breaks, or strips a screw, or whatever.

2h agoHN ↗

Yes, but hardly ever at Astra, because it usually gets the job done. Definitely at worse models when you give them very clear instructions and goals to follow and they respond telling you that the job has been completed perfectly, verified byte-identical, all tests pass, etc etc and you look at it and it’s a pile of garbage.