[scubbo]

> The bot does the trivial things well

Clearly not, since all the answers were incorrect; two of them by an order of magnitude:

* 0.93((1.03)^1000) is 6.393E12, not 1.788E13

* 0.93((1.02)^1000) is 3.7E8, not 4.28E9

* 0.93*((1.01)^1000) is 19,492, not 19,482

...on a whim, I just tried asking ChatGPT "What would 93 cents accumulate to over 1000 years with 3% compound interest?", and the answer (179.74) was staggeringly wrong because it thought that 1.03^1000 was approximately 193.48.

How timely that https://news.ycombinator.com/item?id=42484937 was posted today.

[dgfitz]

To be fair, I qualified the whole thing with “I asked a bot” so I can say I learned a valuable lesson as to the value of LLMs.

I doubt anyone reading this will believe me, today is the second day I have ever tried using an LLM. Talk about a backfire.

[scubbo]

Personally, I can say that I just gained a whole _ton_ of respect for you for your ability to learn a lesson, to admit that you made a mistake, and not to double-down on insistence that LLMs are Good, Actually.

I hope my message didn't come across at too unpleasantly confrontational - I'm not annoyed at _you_, but rather at the over-reliance of these hallucination machines in our industry which is supposed to prize hard data and accuracy. I'm glad I was able to help someone gain a bit of reasonable skepticism for them!

All the very best to you and yours for this holiday season!