Hacker News Reader

HNHacker News

Top New Best Ask Show Jobs

Comment by TradingPlaces | Hacker News Reader

Full thread

TradingPlaces·

Already a BS machine for me on first try.

Me: Can you manipulate data tables?

C2: Yes I can. Here’s some of the things I can do.

Me: Here’s some data and what to do with it (annualized growth rates).

C2: [processes for a while and starts spitting out responses, then deletes all that]

Me: What happened?

C2: Sorry, I lied. I can’t do any of that

Full exchange: https://econtwitter.net/@TradingPlacesResearch/1106958439180...

View on HN

Wait, that actually sounds wonderful! This is the second best option of what happens when you have an over eager assistant: they try to help and then notice they are out of their dept, so they let me know, before they waste my time.

Could have just said “no” to the first question, saved me time, and Anthropic GPU inference compute, which adds up quickly.

But as I noted elsewhere, I am finding it very useful for text summarizing.

And to follow up on "Anthropic GPU inference compute, which adds up quickly,” I’ve already been rate limited.

That's amazing. I've been waiting for someone to crack that.

Believe it or not, getting the output "Sorry, I lied. I can't do any of that" is a huge accomplishment.

How much of that is due to the “prompt interpreter” and how much of that is due to the LLM?

Perhaps someone at the factory[1][2] stepped in.

[1] https://www.theverge.com/features/23764584/ai-artificial-int...

[2] https://time.com/6275995/chatgpt-facebook-african-workers-un...

TBF, finding it very useful for text summarization

Reply on news.ycombinator.com