A bunch of sections of the README are nonsensical as well. I'll just file this under "magical token-saving project #481 of 2026 that probably performs worse than the harness" for now.
Feels like it to me too. Not only is it clearly extremely vibecoded, but the human author seems to have failed at (or not tried) restraining common agent impulses such as writing huge, sprawling stream-of-artificial-consciousness "documentation" which is of no use for humans, and questionable use for the agents themselves.
My policy is that if someone doesn't understand the tool they've vibe-coded well enough to make the documentation coherent, they don't actually use it, and so I probably shouldn't use it either.
Opus can't speak coherent English anymore. I dunno what happened, but it's not speaking a language humans can understand. It uses unusual shorthand and slang, which it seems to use pretty consistently, so it's probably possible to learn its particular dialect. But, I'd rather it speak a language I already know.
That doesn't really track, though. Even tiny models, like Gemma 4, have coherent prose. I know Opus isn't being quantized that small. It seems like it must be some kind of...I dunno. Maybe over-fitting toward some user metric that doesn't track how readable its writing is?
It seems to still be good at code (though I haven't directly compared to earlier Opus versions lately), so it's not a general model collapse type problem, nor quantization errors. If quantization problems, I would expect it to break down on logic before prose, since a 4-bit quantized Gemma 4, even the small versions, have pretty good and, more importantly, coherent written English.
This repo is amazing, I coded all weekend w fable w the orchestrator model and asked fable how much it helped, half the tokens and essential eliminated work w what quality checks
What's with slop-factories forgetting what a README is supposed to be? A quick intro to the project, not a gigantic abomination with irrelevant details.
From the hero image:
What on earth does this mean?A bunch of sections of the README are nonsensical as well. I'll just file this under "magical token-saving project #481 of 2026 that probably performs worse than the harness" for now.
Feels like it to me too. Not only is it clearly extremely vibecoded, but the human author seems to have failed at (or not tried) restraining common agent impulses such as writing huge, sprawling stream-of-artificial-consciousness "documentation" which is of no use for humans, and questionable use for the agents themselves.
My policy is that if someone doesn't understand the tool they've vibe-coded well enough to make the documentation coherent, they don't actually use it, and so I probably shouldn't use it either.
Opus can't speak coherent English anymore. I dunno what happened, but it's not speaking a language humans can understand. It uses unusual shorthand and slang, which it seems to use pretty consistently, so it's probably possible to learn its particular dialect. But, I'd rather it speak a language I already know.
I assume all major cloud AI are lowering their quants and measuring user retention.
That doesn't really track, though. Even tiny models, like Gemma 4, have coherent prose. I know Opus isn't being quantized that small. It seems like it must be some kind of...I dunno. Maybe over-fitting toward some user metric that doesn't track how readable its writing is?
It seems to still be good at code (though I haven't directly compared to earlier Opus versions lately), so it's not a general model collapse type problem, nor quantization errors. If quantization problems, I would expect it to break down on logic before prose, since a 4-bit quantized Gemma 4, even the small versions, have pretty good and, more importantly, coherent written English.
Readme is very AI.
Still will see what inspiration I can take from this. I’ve seen improvements from my own rudimentary repo overview tool
This repo is amazing, I coded all weekend w fable w the orchestrator model and asked fable how much it helped, half the tokens and essential eliminated work w what quality checks
> task-shaped skills
What's with slop-factories forgetting what a README is supposed to be? A quick intro to the project, not a gigantic abomination with irrelevant details.
Is at okay to flag things if they just feel like end-to-end slop? If this was a comment, I'd down-vote it.