Hacker News Reader

Comment by CuriouslyC | Hacker News Reader

_By default_ GPT gives the average advice. You can prompt it to take perspectives that will lead it to give very different advice.

View on HN

I find it follows wherever you lead it, but if I know what kind of advice I need, what information is the bot adding?

For me it's a form of rubber-duck debugging.

I've never managed to make it work with an actual rubber duck. Talking to an inanimate object doesn't make me think any differently, but there's something in trying to formulate things to get the LLM to "understand" my problem that triggers the same brain mechanisms I get when explaining it to a real person.

At the end of the day, the advice seeker is responsible for the evaluation of given advice for their situation.

That's a key difference between advice and instruction.

The advice giver cannot know the seeker's situation completely, nor can the advice seeker know the giver's.

An AI cannot be trusted to give The Answer. But you can use AI to explore other perspectives and critique your thoughts—not that its critiques will be correct either. Nonetheless it can help get you on different thinking paths that your normal thought patterns wouldn't guide you on.

When I was a kid and needed to know how to spell a word, my teachers told me to look it up in a dictionary. This had the problem you're describing — if I could do that, I wouldn't need to.

But AI? I can ask LLMs questions like:

"A couple is buying a house in Germany. What questions do they typically forget to ask, which they regret not asking, and often wish later that someone had told them to ask?"

And it can fill in the gaps better than when I ask a human the same kind of question.

I don't expect it to be perfect, this could well be a half-arsed boilerplate, but the actual humans who I asked this of, mostly responded "Huh? I don't understand?"

https://chat.openai.com/share/f17b013e-6b24-4b49-9986-78a5e4...

(Note this chat had been given custom instructions, that's why it's responding with this unusual pattern).

Not necessarily, sometimes you need it to give you writing advice from the perspective of Hemingway or interpret someone's motives from the perspective of Jane Austen.

Isnt that an extension of the same problem though? You probe it to take the statistically most accepted alternative perspectives that exists in its data. Its not intelligent in the sense that it comprehends what you or it are saying, it just follows the weights.

It definitely doesn't just "follow the weights". It can follow instructions that go down paths where there would be no similar training data and no simple completion that could result in that response.

What it does probably isn't "comprehension", but we also have no idea what comprehension is so it's not a very good target.

Reply on news.ycombinator.com