ScubaHogg wrote: Sat Jan 18, 2025 11:55 pm
One fun prompt for me has been when I want an intro to a new subject. Whatever it is. I just ask for ChatGPT to explain it to me like I’m a 10-year old. It’s been a nice way to get a high level summary. Or I had some questions regarding electricity. I asked it to explain it to me like I was 10, using a river as an analogy to electricity.
Same. I've asked it for things in my formal field of knowledge, and it answered perfectly correct, in the way a topmost expert would. And it does it instantly, and arguably more concisely. I've asked it things that I don't know, like astronomy, and the explanations are satisfying as if I was asking a public-educator like Brian Cox.
snic wrote: Sun Jan 19, 2025 7:05 am
And how do you know that ChatGPT's "explanation" wasn't just a hallucination?
...
It's not a question of "should I learn to use AI?". I'm willing to use any tool that makes life easier. My concern is whether I can trust it. If I can't trust it, I have to verify its output by doing more reading and analysis. But then how has AI made my life easier relative to using other tools, such as internet searches, forums like this, etc?
The same way you might challenge a subject matter expert in real life. You wouldn't blindly trust me on any given answer if I told you I was a credentialed [biologist/chemist/meteorologist/zoologist] would you?
You (1) use your intuition (2) ask for further justification (3) request references and sources. By default it already breaks the answer into logical piece-wise chunks, and often provide math equations when relevant.
At the very minimum, it gives you a shortcut to the answer. If I wanted to know, for example, the real estate capital gains tax rate in Mexico, I could google it, and find the answer buried in hundreds of blog articlse of private accounting firms, of discussions on message board, and some outdated guide by ernst and young, or laboriously translate my way across the gigantic tax code of the govt in its native language.
Or I could ask the LLM to cite the exact location of the text and then read it myself.
Better, I can tell it the hypothetical transaction prices, dates andd geography (if municipality and holding period matters), offsetting capital transactions, and have it
chain and calculate the final answer by itself. Again, all these things a reasonable person can eventually find and verify on their own, but its a great accelerator.
Wooly questions of subjectivity on esoterica is one thing (though it'll answer that too), but for many topics for which there is a large corpus of information, it converges to the "correct" answer remarkably well.