Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Show HN: I RL-trained an agent that trains models with RL (for ~$1.3k)

github.com

Post reply on HN
Print view View original

31–40 of 54 posts

Re: Show HN: I RL-trained an agent that trains models with RL (for ~$1.3k)

#31

Post by amelius »

callamdelaney wrote: post #6

AI trains AI already, agents are happy to spin up real training pipelines for deep learning or regression models or whatever you want right? I guess the advantage to your project is that it provides a framework to allow the agent to access extra compute?

I guess more important than LLMs training LLMs (which is mostly the same code over and over) is LLMs cleaning and curating training data.

Re: Show HN: I RL-trained an agent that trains models with RL (for ~$1.3k)

#32

Post by piratebroadcast »

r_lee wrote: post #19
Danau5tin wrote: post #5

Earlier quoted context omitted.

Mainly Fable, but It was me who wanted to emojis added hah. I also of course edited the README by hand (crazy I know), but the code is entirely fable

why...?

To circumvent distillation.

Re: Show HN: I RL-trained an agent that trains models with RL (for ~$1.3k)

#33

Post by throwa356262 »

callamdelaney wrote: post #6

AI trains AI already, agents are happy to spin up real training pipelines for deep learning or regression models or whatever you want right? I guess the advantage to your project is that it provides a framework to allow the agent to access extra compute?

Not sure how much of it was marketing but M2.7 supposedly trained itself

https://www.minimax.io/blog/minimax-m27

Re: Show HN: I RL-trained an agent that trains models with RL (for ~$1.3k)

#36

Post by Danau5tin »

rolls-reus wrote: post #29
Danau5tin wrote: post #5

Earlier quoted context omitted.

Mainly Fable, but It was me who wanted to emojis added hah. I also of course edited the README by hand (crazy I know), but the code is entirely fable

i read around launch that anthropic will fallback to opus if fable is used for frontier LLM development. did you run into anything like that?

Strangely, I did not. I was expecting it and looking out for it the whole time. At least I never say a warning!

Re: Show HN: I RL-trained an agent that trains models with RL (for ~$1.3k)

#37

Post by irickt »

saberience wrote: post #14
babelfish wrote: post #13

Earlier quoted context omitted.

Did you read the README?

The AI generated README?

Your comments are casting aspersion without showing that you have looked into the work or its author. This project is a case study in good work done with heavy ai assistance while it is clear that there is a skilled person leading. I predict that this will become very common and welcome here.

Re: Show HN: I RL-trained an agent that trains models with RL (for ~$1.3k)

#38

Post by bitbasher »

netvarun wrote: post #4
bitbasher wrote: post #3

Lots of emoji in that readme. Was it mainly codex?

“The Anthropic team for their incredible coding models (Fable-5 wrote every line of code in this project), and the Claude Code harness.” Source: the repo

Funny how I could tell without reading anything. I immediately lost interest upon noticing.

Re: Show HN: I RL-trained an agent that trains models with RL (for ~$1.3k)

#39

Post by Danau5tin »

throwa356262 wrote: post #33
callamdelaney wrote: post #6

AI trains AI already, agents are happy to spin up real training pipelines for deep learning or regression models or whatever you want right? I guess the advantage to your project is that it provides a framework to allow the agent to access extra compute?

Not sure how much of it was marketing but M2.7 supposedly trained itself https://www.minimax.io/blog/minimax-m27

Very cool! Thank you

Re: Show HN: I RL-trained an agent that trains models with RL (for ~$1.3k)

#40

Post by Danau5tin »

amelius wrote: post #31
callamdelaney wrote: post #6

AI trains AI already, agents are happy to spin up real training pipelines for deep learning or regression models or whatever you want right? I guess the advantage to your project is that it provides a framework to allow the agent to access extra compute?

I guess more important than LLMs training LLMs (which is mostly the same code over and over) is LLMs cleaning and curating training data.

Very good point. Sifting through noisy data and creating curated datasets could be of great value. Perhaps worth a project by itself!
Post reply on HN