‹ BackHN Continuity

Thread

Bend – a language that blocks AI mistakes via proof and runs on GPUs

616 points · 327 comments · nicolas-siplis

  1. shantnutiwari · · focus · HN ↗
    Yeah, but this assumes the llm will follow the "Laws". I find llms routinely ignore steering docs etc, even outright instructions. Like "Dont use python", next line it is trying to use Python.

    Seems to me the llm will just try to work around the "laws"

    1. serial_dev · · focus · HN ↗
      Exactly, you can have all the laws, but an LLM (or even a human) will just delete or refine the laws and hide the change in a 5kloc PR, that the "code author" will not review carefully, and you have 5 min to review and stamp to make sure your company is moving fast...
      1. andy12_ · · focus · HN ↗
        If the LLM changes the laws to bypass them that's on you. The whole point of this is that you don't have to manually review most code written; only the laws. If the LLM changes the laws and you ignore it that's a you problem.
        1. serial_dev · · focus · HN ↗
          I totally get it, but in companies where you gotta crunch out 10x more features, review 10x from other people, this could be easily overlooked.
Open on Hacker News to reply ↗

Unofficial Hacker News client; not affiliated with Y Combinator.