CODA now has a released page.

[AndrewGrant]

the agent produced a genuinely strong, original engine

:DDDDDDD???

What engine do you think CODA is based on?

Coda is a derivative of the following engines:

Alexandria, Altair, Arasan, Astra, Berserk, BlackMarlin, Caissa, Cinder, Clarity, Clover, Halogen, Hobbes, Horsie, Icarus, Igel, Integral, Koivisto, Midnight, Minic, Motor, Obsidian, PlentyChess, Quanticade, Raphael, Reckless, Rubichess, Seer, Starzix, Stockfish, Stormphrax, Tarnished, Tcheran, Texel, Tucano, Uralochka, Velvet, Viridithas, Wasp, Weiss, Winter -- as well as Ethereal, and any other since-removed references, and every single engine that Anthropic has provided to Claude in training.

As such, Coda is obliged to comply with the license attached to every single one of those projects. GPLv3, AGPLv3, MIT,and possibly others.

An LLM is not a clean-room reproduction from a specification. It is a regurgitation of stolen assets.

[Ciekce]

What engine do you think CODA is based on?

https://github.com/adamtwiss/coda/tree/ ... gine-notes

as a side note, do you not find it extremely disrespectful to paste a wall of LLM slop to your human interlocutors? I know I do

[chrisw]

What engine do you think CODA is based on?

https://github.com/adamtwiss/coda/tree/ ... gine-notes

as a side note, do you not find it extremely disrespectful to paste a wall of LLM slop to your human interlocutors? I know I do

I’m not a part of your ecosystem, so I lack general informative gossip and so on that goes on within it. But, looking at what you posted very quickly, seems he had an engine, he asked Claude to enumbrate differences between a mass of GitHub GPL engines, for novel ideas, asked the AI to implement the idea and then tested it, presumably SPRT. Ideas that worked were kept and the process iterated through all engines he could find.
The LLM would not have “copied” the code any more than you copy Shakespeare. It kind of works it out, but that’s a philosophical question way beyond chess programmers pay grade.
Then it’s published with the GPL licence, like almost everything else. Has he done anything illegal? Has he shown a novel creative approach? Has he tried to deny anything? What’s your gripe?

Btw, I find it a good approach to use AI for chess engine analysis as per my work over the last couple of days. It picks up on way more stuff than a human can and it compares (I am assuming) without bias for or against any engine or programmer. All get the same treatment. Probably impossible if we asked involved humans to do the same thing. Yes, I share your concerns about AI slop but there are cases, many cases, where AI is extremely useful and effective.

[smatovic]

El Futuro

--
Srdja

[mar]

El Futuro

a future of what? LLM is not a road to AGI (do we even need AGI?)
slopmachines only parasite on decades of human knowledge and they're even completely unreliable
at what they "produce"

recursively training LLMs on slop degrades the quality to the point where it cannot even compose
intelligible sentences anymore, hence LLMs heavily depend on the quality of the input they were trained on
https://www.nature.com/articles/s41586-024-07566-y

I have the feeling that only the incompetent praise slop, since we live in an era where (sadly) mediocrity
and incompetence is being normalized, this is bad news for the future

[adamtwiss]

Thanks for the interest — and for the pushback. Both are welcome; a thread like this is part of how you work out where the lines actually are in fairly new territory.

I've never hidden what Coda is: it's agentically developed, top to bottom — it's in the tagline. Whether that excites you or annoys you, it's the direction the field is heading, and I'd rather explore it in the open — design docs and the full experiment log committed to the repo — than ship a black box. The whole thing is GPL. The intent from day one has been to give everything back, not to take.

Like every engine author here, I've studied what works elsewhere and tested it — that's how this field has always moved. Ideas and algorithms generally aren't copyrightable, and where a specific idea came from a specific engine I've tried to say so; the repo is full of those credits. There's original work in Coda too — the x-ray threat inputs, for one — so "it's just other people's engines" doesn't quite land either.

That said, I take licensing seriously and I don't want to get it wrong. If you have a specific concern, I genuinely want to hear it, and I'll deal with it directly rather than argue it.

Which brings me to Reckless, since that's the one raised. It's a great engine, we've studied it closely, and it's also the one engine we lean on that's AGPL rather than GPL — the case where the details genuinely matter. So I spent today auditing exactly that, and turned up a few narrow spots that weren't as clean as I'd like — some adapted code, and a couple of small tables that should have been re-derived more independently or credited more clearly. Real, but narrow. I'd rather fix them than argue them, so they'll be sorted over the next few days — out in the open in the repo, like everything else.

Appreciate everyone taking the time. Honestly, the thread made Coda a little better today.

(And yes — the "we" here is Claude and me, today's audit included. An irony I'm well aware of.)

[Sylwy]

An AI (ChatGPT) analysis of the opinions in this thread:

[Rebel]

Did a quick Coda 0.90 match against a 3578 elo pool with 10 engines. After 3600 games Coda scored 63.4% which means a rating of 3700 elo.

Welcome to the new future.

[chrisw]

Did a quick Coda 0.90 match against a 3578 elo pool with 10 engines. After 3600 games Coda scored 63.4% which means a rating of 3700 elo.

Welcome to the new future.

yes, I wondered if this was why some people had turned on the programmer, it did seem very personal - like he worked with them and then a possible superstar suddenly emerged. Been here before, no?

[Gabor Szots]

Do AI made chess engines make sense? They are not going to surpass existing top engines, I think. To do that some new idea should be introduced which they can't do.
So this seem to me simply as a show-off.

And, how will I be able to tell if a 'new' engine is AI based? If the 'author' is not honest I won't be able to distinguish a human made engine from an AI made one.