Comment on Well... Are We Rooting For AI to Destroy Jobs?

• SJ0@hilariouschaos.com ⁨9⁩ ⁨months⁩ ago

  Tough to say. Most LLMs are shite at coding too, and that’s much more mechanical

  source

  • Owner_of_donky@hilariouschaos.com ⁨9⁩ ⁨months⁩ ago

    Yes, exactly. Coding is much more technical. Law is just like normal text. If you have a news article and ask it to take the most important stuff from it and bring it back to you why wouldn’t it be able to do the same with laws?

    source

    • SJ0@hilariouschaos.com ⁨9⁩ ⁨months⁩ ago

      There’s a number of reasons.

      First, law is technical in a different way than coding. The classic joke of Bill Clinton asking what the definition of “is” is does have a basis, that when you get into incredibly technical definitions.

      Second, Stare Decisis is a thing which means that you don’t just need to know what’s in all the laws and all the cases, you need to know what precedent supercedes or supplements each other decision. While it is rare, there’s time it occurs that some case from the 1600s common law ends up applying in court. It’s like finding an old opcode in an 8086 manual – can you still use it? What can you surmise about it?

      On a technical level, the biggest benefit of LLMs becomes a weakness because there’s so many things that look like they could be the right answer but aren’t.

      I know one question I asked ChatGPT (which we’ve established sucks at this, but hear me out) was about hazard gas monitoring in industrial environments. I asked about what particular laws applied to it. It created good looking fake citations from whole cloth and even wrote out entire laws that don’t exist, but even if it could prevent itself from doing that, there’s a logical leap you need to take to answer the question – In reality it’s likely covered under the lines in occupational health and safety acts that say something like “employers must do everything possible to ensure a safe work environment”

      source