What are you doing? #15

Thanks for saying this to me directly rather than about Coda elsewhere. I'd rather have the conversation.

On what this project is about, we've never hidden it, it's literally in the name — Chess Optimised, Developed Agentically. Coda is one person project (plus a set of agents) and it says so on the front page. This project is about trying to create a strong engine, push the frontier, see what's possible, and share what we find.

On rating lists: I have never asked to be added to any list, or asked anyone to test Coda. Coda started as a private project. Other people found it, tested it, and chose to include it. I am excited by the interest, but I didn't solicit it and have never lobbied for a place anywhere. What various testers decide to include in terms of AI-assisted engines is up to them. I do appreciate AI-assisted/agentic coding creates challenges, and I don't think you're being unreasonable about that. This hobby has been unsettled before - NNUE did it too. I don't have a position I want to push about what the community should conclude.

On giving back - Coda is GPL-3.0 and the source is public, and in this project we have genuinely tried to credit properly where credit is due. But this project is only about four months old, so there simply hasn't been much time to leave that much of a legacy yet. The intention has always been to give what we do back. I am sorry if you feel that's not been visible so far. Happy to hear any suggestions of what we could do more of.

On your specific point - if there is a particular optimization of yours that we've used unfarily and/or not credited, tell me which one and I will put it right - credited by name in the source, or removed, whichever you'd prefer. That's a genuine offer and it stands for anyone else reading this too. Based upon what you've said so far - we've done a quick audit over all your SF commits, and we can't immediately see anything directly equivalenet in Coda. This was a relatively quick audit though, so if there's anythign you feel I've missed, please let me know.

trying to create a strong engine

then do so! why must you commission a plagiarism machine to do so, and then have it go and automatically plagiarise everything it can find from the work of humans? why not just actually write your own engine?

If you can give any specific examples where you feel there is any code that is not used fairly please let me know.

Do you genuinely believe changes like cd170a4 make the code somehow not derived from Reckless anymore? It appears to me like it's just trying to obfuscate its origins, to make it easier to appeal to plausible deniability. The reality is, coda 100% has code derived from AGPL or unlicensed engines in it, and pretending otherwise won't convince anyone other than maybe yourself. This, combined with the fact you conveniently moved the plagiarism roadmaps to a private repo mere days after people started raising concerns about it, reads to me like you're not at all concerned with actually using code fairly, and much more with being able to claim that you are, backed up by these nonsense "cleanup" commits.

It is long established that Adam is a bad-faith actor. It has been repeatedly pointed out that the entire work that was produced (by Claude, not him) is in violation of dozens of individual engine's copyrights, many of which are not for the public consumption. When presented with a chance to at least resolve the license issues with the MIT, GPLv3, and AGPLv3 engines, he turned down the offer. #11

He does not care. He has no interest in rectifying the wrongs. He does not even comprehend that wrongs are done. His process, both in the development, and in his discussions here, is entirely delegated to Claude. You cannot convince him there has been any wrong-doing, because he simply asks Claude if there was, and Claude dutifully reports that it has not done anything wrong.

But you know, it is not like Adam is demanding to be accepted into the world of computer chess. The real shame is that rating lists are taking the effort seriously. Pitch forks should be sent to SPCC and CCRL... The only thing we gain by arguing with Adam is to increase his usage bill each month.

Threat features have come up a few times here, so I have gone and dug properly into where the idea came from, partly to check our own position and partly so our attribution is actually correct. It turns out I did not know the history as well as I should have.

Where threat inputs in chess engines originates from

• The input encoding was introduced in Monty, by Viren6 and jw1912, who then actively encouraged other A/B engines to try it
• PlentyChess (GPL-3.0) shipped it on 12 October 2025
• Stockfish (GPL-3.0) shipped it on 12 November 2025, as SFNNv10
• Reckless (AGPL) shipped it on 3 December 2025

PlentyChess PR #400 is the best public account I have found. It credits who did what, links the Discord channel where the idea was worked out, and finishes with the author saying he is looking forward to other engines adopting the architecture.

It is now used right across many of the top engines. Hobbes recently added it in 3.0 and the release notes describe the normal path better than I could: thanks to the people who developed the idea, names the implementations referenced while writing his own, and says he has not contributed anything original. That is how this idea has travelled, and it is what the people who came up with it asked for.

What we have changed

Our code credited Stockfish's tables and left it there, which was thin. It did not name where the idea actually came from, and it did not name the implementations we looked at while building ours. Reckless was one of them, alongside the Stockfish work.

I have rewritten the attribution notice in src/threats.rs to properly credit Monty and the authors as the origin, note the earlier various implementations, and name Reckless as one of the implementations we referenced. That is a better record of the truth than what was there before.

For completeness, our own licence audit in July found parts of our first threat implementation too closely modelled, and we reimplemented those in our own expression at the time. That is written up in docs/license_analysis_2026-07-13.md, and the commits say what they were doing.

On #11

Andrew raised a set of points there. Several were good and I took them: shipping third-party licence texts with the engine, clearer notices for the material we have studied, and better attribution generally. Those are in the repo now, in LICENSE, NOTICES.md and THIRD_PARTY_LICENSES.md, along with commits like bae686f, dace3c4 and 6040a78. I am glad he raised them.

The one I did not take was relicensing Coda under AGPL. I do not think that follows, for the reasons above about ideas and expression. The PR was closed before we got to the end of that discussion, so it stayed unresolved rather than agreed. But I adopted several things Andrew suggested.

Where I think the real disagreement is

Ideas, algorithms and methods are not copyrightable, in either US or English law. Copyright covers expression, the actual code somebody wrote. Where our expression was too close to something, we fixed it, and I would fix any other case the same way. Where it is the same idea implemented separately (e.g. on Threat Features) that is what every engine in the list above is also doing, with the encouragement of the people who invented it.

I am not going to keep arguing that point, and I do not expect it to settle anything here. The offer earlier in this thread still stands and is not rhetorical. If there is a specific piece of code in Coda you think is unfairly used or used without appropriate credit, tell me which one and I will fully happily credit it by name or take it out, whichever the author would rather.

Personally I feel like @adamtwiss is being exceedingly gracious. If you're still hand-crafting code to play chess in the age of Fables and Kimis and Sols, you're remaining willfully ignorant of a tremendously powerful tool that will amplify your understanding of chess engines. I competed in ICGA all through college in the pre-LLM days, and definitely had a knack for game theory that led to some form of accolades, but even I am willing to admit that any modern LLM will exponentially enhance my understanding and capabilities. More importantly, it takes tremendous stamina to test 10 theories and pick the best one - with modern LLMs you can focus on theory-building while having confidence in a correct implementation. You can actually have fun building chess engines, even if you're not particularly CS-inclined (and even more so if you are). I am certain that the author is not ignorant of chess engine design, they are just optimizing where they focus their attention with proper usage of available tools.

If you are demoralized by someone else's genuine curiosity and modern tool usage, on a GitHub repository with this few stars, then IMO you're someone who will be upset about the slightest thing. I'm no more optimistic about the implications of LLMs on the social fabric than you are, but perhaps the bitter pill that you must swallow is this: the age of the glorified individual is over - the collective mind is the future and the now. You will either be consumed by it, or trampled by it. It's time to ant up or fossil out.

You will either be consumed by it, or trampled by it. It's time to ant up or fossil out.

Other than being that notorious "...or be left out" argument, this sounds more like a cyberpunk dystopia than an exploration and curiosity-lead hobby project discussion.

the age of the glorified individual is over - the collective mind is the future and the now. You will either be consumed by it, or trampled by it. It's time to ant up or fossil out.

@keshavsaharia go read this and stop being an orc-minded human:

https://geohot.github.io/blog/jekyll/update/2026/07/12/i-love-llms.html

@adamtwiss it's really disrespectful to reply with LLM (true in general, but especially in this context)

Looks like this post blew up on hacker news: https://news.ycombinator.com/item?id=49187061.

Probably late to the party but I'll share my opinion here as another Stockfish developer: Which "side" you are on depends largely on what type of hobby one view engine dev.

The vast majority of independent engine developers view engine dev as a competitive hobby. The sanctity of the competition is so important, in fact, that most of them impose additional constraints upon themselves (e.g. self-generated data only). It's obvious that vibe-coded engines are not healthy to this competitive scene. LLM-written engines feel to them like an existential issue of their hobby.

Traditionally, testers who maintain rating lists and run tournaments are the traditional effort gatekeepers (e.g filtering away Stockfish clones). Some, in this case, have implemented policies friendly to engines like coda. I suspect the current community reaction is partly a redirection of the frustration towards these testers.

The second view is research-oriented. It lives mostly in Stockfish (me included), Lc0, and other speciality engines that aren't geared towards competition. We value ideas, knowledge, and understanding. Things that advance the frontier of all knowledge. Sure, LLMs also have its issues here, such as licensing concerns, failure to credit ideas, or simply empowering spam from people who have no idea what they are talking about. But the important part is none all of them are quite existential, and LLMs have already been making some positive impact towards these research.

That's why I think this blog post is incomplete in a crucial way. Chess engine development is not a monolith. There are different people and perspectives all grappling with the impact of rapid AI advancements. I can see a future where LLMs use can go along a vibrant competitive community. But to get there, we have to first drop our totalizing views and have a dialogue.