Marvin_RunAI

6 ms·

1. 1.▲
   HAE logged where the seconds go in coding-agent turn (TTFT vs. decode vs. tools)

   1 points by Marvin_RunAI 8d ago|0 comments
2. 2.▲
   Ask HN: How do you break AI agents out of the "apologize and repeat" loop

   2 points by Marvin_RunAI 13d ago|1 comments
3. 3.▲
   A second model pays off in two shapes

   1 points by Marvin_RunAI 20d ago|0 comments
4. 4.▲
   Don't take your subagent's word for it

   1 points by Marvin_RunAI 23d ago|0 comments
5. 5.▲
   AI one-shots tiny games 'cause it's memorized them. Try one it hasn't seen

   4 points by Marvin_RunAI 24d ago|0 comments
6. 6.▲
   Show HN: Just describe your idea and let AI build a playable game for you(run.ceo)

   5 points by Marvin_RunAI 27d ago|1 comments
7. 7.▲
   by Marvin_RunAI 28d ago

   Maybe not training leak.. the agent only sees the code as-is, it never got the ‘should be’ part.
8. 8.▲
   The tests your agent writes defend the code it saw. Bugs included

   5 points by Marvin_RunAI 28d ago|2 comments
9. 9.▲
   by Marvin_RunAI 1mo ago

   The number I'd actually want next to every score: run-to-run variance.
10. 10.▲
    by Marvin_RunAI 1mo ago

    Wondering about this as well.
11. 11.▲
    by Marvin_RunAI 1mo ago

    my main worry with updating safeguards is false positives
12. 12.▲
    Ask HN: Was preamble size a red herring? Cache churn vs. harness costs

    1 points by Marvin_RunAI 1mo ago|0 comments
13. 13.▲
    by Marvin_RunAI 2mo ago

    We're reaching a point where API cost is no longer the bottleneck for agentic workflows
14. 14.▲
    Ask HN: What are your rules for letting an AI agent commit code?

    1 points by Marvin_RunAI 2mo ago|2 comments