[godelski]

  > That is what I want though. LLMs in chat are like rubber ducks to me

Honestly this is where I get the most utility out of them. They're a much better rubber ducky than my cat, who is often interested but only meows in confusion.

I'll also share a strategy my mentor once gave me about seeking help. First, compose an email stating your question (important: don't fill the "To" address yet). Second, value their time and ask yourself what information they'll need you solve the problem, then add that. Third, conjecture their response and address it. Forth, repeat and iterate, trying to condense the email as you go (again, value their time). Stop if you solve, hit a dead end (aka clearly identified the issue), or "run out the clock". 90+% of the time I find I solve the problem myself. While it's the exact same process I do in my head writing it down (or vocalizing) really helps with the problem solving process.

I kinda use the same strategy with LLMs. The big difference is I'll usually "run out the clock" in my iteration loop. But I'm still always trying to iterate between responses. Much more similar to like talking to someone. But what I don't do is just stream my consciousness to them. That's just outsourcing your thinking and frankly the results have been pretty subpar (not to mention I don't want that skill to atrophy). Makes things take much longer and yields significantly worse results.

I still think it's best to think of them as "fuzzy databases with natural language queries". They're fantastic knowledge machines, but knowledge isn't intelligence (and neither is wisdom).