
Cody Blakeney
21.8K posts

Cody Blakeney
@code_star
Leading research at @arcee_ai | Formerly Data Research Lead @DbrxMosaicAI | Visiting Researcher @meta | Ph.D | #TXSTFOOTBALL fan | https://t.co/4G6Jf3b0V4



Another thought here (and man I feel old saying this). Is that people tended to solve the problem this solves in the language they were most comfortable with in the past. Yes, I have read the replies, and I’m very happy for you people that wrote lots of bash. I don’t think it’s out of line, to say unless you worked on infrastructure, or were a strange hardcore Linux guy, you probably didn’t do all that crazy of bash other than what you new for scooting around the terminal. You could just use like .. a real language for reliability. Also, if you were a REAL Linux nerd you probably were dabbling with something like fish instead of bash before 2020 and only switched back to zsh/bash because there were no docker images setup for fish and wasn’t going to work with GPUs.




This is actually incredible. We are living in the post-smoke test software era.

This is actually incredible. We are living in the post-smoke test software era.

It’s crazy because it seems like it’s significantly more emergent as a result of LLMs than even the most famous LLM word.

This is actually incredible. We are living in the post-smoke test software era.


You think I’m joking?


You think I’m joking?


What’s funny about this, is it’s not evidence no one knew the command before this. It’s just a marker of when that exact set of flags, in that order, made it into someone’s post-training dataset.


the rather funny thing about LLMs is that knowing how transformers work teaches you literally nothing about why LLMs do what they do, or how to use them better










