> But the reliability problem is being fixed quickly, as researchers create models that can check their own work and improve themselves. Soon they will be substantially better than humans at many tasks.
Yes, no.
They don't. They just push probabilities higher. The tech is non-deterministic.
Humans are also non-deterministic, for all practical purposes, and the more of them you need to build some piece of software, the harder it can get to keep the likelihood of issues low.
I'm not saying it will happen, but at least it's not inconceivable that LLMs could surpass them in this calculation more and more often.
My interpretation of the argument is "Humans are also non-deterministic, but we make organizations work on a daily basis despite that. We could do the same with LLMs, which may in fact turn out to be less non-deterministic than humans."
I agree with both takes; organizations have evolved many error-correcting mechanisms like bureaucracy, laws and contracts to keep a gaggle of non-deterministic actors behaving deterministically. There is no reason those actors need to be human.
why do you think that something being non deterministic means it can't be better than humans at a task? hasn't this same approach been used when developing alpha go which has surpassed humans?
Honestly, I think it will come down to liability. With humans in loop, the answer is obvious. It's the human who made a decision. When you have AI make a decision, who is responsible? The AI provider, the person implementing things?
I was debating about this with my cousin, he's a doctor. Doctors are already using AI, but they are still the judgement maker, If the AI makes a mistake, they and the entity they work for are held liable. But it might take a long time to realize that a mistake has been made, so it's going to take a while to work this out. He was arguing that humans are so fallible that it AI can replace them in his field.
Yes, no.
They don't. They just push probabilities higher. The tech is non-deterministic.