Comment by unsnap_biceps

3 months ago

one that you would trust the lives of thousands of humans to every day? It seems unlikely we are anywhere close to a point where we can ensure that any AI won't hallucinate and cause an issue.

3 comments

unsnap_biceps

Reply
kelseyfrog  3 months ago

Sure, AI can spit out nonsense, and that’s a real concern. But in engineering, we deal with imperfect tradeoffs every day - it’s baked into the job. If we insist on a flawless solution before shipping anything, we’ll never ship. There’s always an optimum where we uphold safety standards without sacrificing forward progress

  • unsnap_biceps  3 months ago

    Yes, however, we need to hold different bars depending on the consequences of the failure. AI in a game that randomly spins around and breaks emersion for a player is fine, an AI that crashes a plane is not fine.

    I have zero issues with research into AI research into those areas, but I think it does a major disservice to claim that a weekend is all it takes to get something close to ready for life or death decisions.

  • ta1243  3 months ago

    Any solution would have to be better than the current one. The last 30 years of the computer industry tells me that is not going to happen.