[jahsome]

I've had numerous days just like this.

The perfect use case for ChatGPT for my personal workflow has been to give me the 70-percent-of-the-way-there skeleton for a given query, Terraform module, or even just to prevent myself from lookup up the syntax of for loops in <language> for the 30,000th time. I can then take that 70 percent chunk and bolt on my edge cases as I go.

There hasn't really been a time where the answer didn't need some massaging and of course occasionally it's flat out wrong, but that's becoming more and more apparent to me those instances are the result of feeding it sub-optimal prompts. I've found it to be eerily similar to my experience learning to use search engines back before SEO titles were a "thing."

It's frankly incredible how little I find myself using a traditional search engine compared to six months ago, at least in the context of work stuff. I'm less inclined to feed it general knowledge prompts, but it's encouraging to see the remarkable LLM tech leaps from generation to generation, and in such short order.

[d4rkp4ttern]

Slightly tangential. After interviewing a candidate, within 20 minutes they sent me a long thank you note that had several generic things that definitely did not apply to what we discussed. Very obviously ChatGPT-generated. Using ChatGPT in and of itself may not be all that bad, but this shows that the candidate did not even take the time to tailor it to our conversation. During the interview I was on the fence about pursuing the candidate but this justifies a clear reject.

[shallichange]

I had it write my green card justification. Awesome corp-speak BS.