HN Explorer

35 comments

[ 4.3 ms ] story [ 45.7 ms ] thread
[dead] Danau5tin
[flagged]
debpalash
Curious whst was your prompt/spec design process. How did u maintain the goal?
bitbasher
Lots of emoji in that readme. Was it mainly codex?
netvarun
“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
bitbasher
Funny how I could tell without reading anything. I immediately lost interest upon noticing.
Danau5tin
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
r_lee
why...?
snerbles
I guess we found the target audience for models that slather READMEs in emojis.
sosodev
The thing we must keep in mind with any of the AI-slop style writing is that it was reinforced behavior because humans wanted it.
rolls-reus
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?
Danau5tin
Strangely, I did not. I was expecting it and looking out for it the whole time. At least I never say a warning!
se4u
that's because they said they'll silently fallback in case of model training. Then they backtracked, and said they wont silently fallback, but how will anyone ever know?
callamdelaney
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?
Danau5tin
Yes I'd heard the labs (Anthropic mostly) speaking about LLMs training LLMs, so I wanted to make things a little more concrete and test it out myself! Essentially you are correct though, my framework allows the agent access to compute, but also the agent itself is being trained to become better at training models with that compute.
vorticalbox
i remeber reading in one of the release blog posts that that version was the "first that codex helped train"
amelius
I guess more important than LLMs training LLMs (which is mostly the same code over and over) is LLMs cleaning and curating training data.
Danau5tin
Very good point. Sifting through noisy data and creating curated datasets could be of great value. Perhaps worth a project by itself!
RamblingCTO
No idea why you got downvoted into oblivion with the context post. Cool idea!
satvikpendem
Because HN detects AI comments and automatically makes them dead. I vouched for it as it's important for this post as context.
bicepjai
Trying to understand, So you downvote the post because of AI comments in HN ?
gnabgib
HN the software, not HN the community (and no.. that isn't what happened)
gnabgib
That wasn't AI detection.. it was self-promotion detection, but hey.. you've helped a promoter out.
saberience
Can you explain how it works?

What problems would it do well on and why?

Where would it start to fail/break?

What are the limitations of a system like this?

When you vibe code a system in a complex area like RL, you basically have zero understanding of what its actually doing, whether its actually any good or not, what you're actually benchmarking, and when the system would fail.

It's the blind leading the blind.

nnevatie
I'm curious to whether the recursively trained models degenerate to troglodytes after a couple of generations.
[dead] no7z
[dead]
tantalor
I think that En Dash is supposed to be Tilde

  –$1.3k -> ~$1.3k
UltraSane
I want to see a latent space of all neural network weights where each point represents an entire neural network
azertyvode
How do you prevent the agent from reward-hacking the hidden eval? e.g. writing training data that effectively leaks the eval distribution rather than teaching a general skill?