[–] 1 point 4 months ago

No. I most often see the AI generate python to do things it can’t directly. But this is a huge question: that program could do anything, do you let it?

How do you safely use an AI, anyway? You start with guardrails, like limiting it to read-only, you tell it to plan only, make sure that any data is also held in a safe place, when it prompts you to let it run something, try to understand it first ….. but in order to get any benefit from AI, you have to let it do potentially dangerous things.

  • source
  • There's no conscious goal to biodiversity. But over time, life will grow and reproduce to harvest as much of the universe's energy as possible and make as many copies of itself as possible.

    Think bacteria on a petri dish, humans on the earth, and really, all living material in the universe.

  • source
  • [–] 9 points 4 months ago

    From an AI safety perspective, the language used isn't that important. An analogy might be asking what kind of hammer is good and safe for building a bomb. Some hammers might let you get the job done easier and/or faster, but ultimately you can do it with any of them (and get into long arguments with hammer fans on which one is the best). Similar to the hammer example, "powerful" isn't a meaningful quality either; is a hammer with a nail extractor thing more "powerful" than a heavier one you can swing harder?

    There is lots of nuance, of course, but it comes down to "best tool for the job" rather than one being more "powerful" than another.

  • source
  • [–] 23 points 4 months ago (2 children)

    Programming languages are pretty interchangeable at the heart of it, so you're not gonna get consensus on "most powerful," and no, it's not safe to use with ai because llms are not healthy and safe for humans and other living things.

  • source
  • view more: ‹ prev next ›