Hi ! I noticed you have attracted upon yourself quite the unfair criticism, the kind of which is seemingly written by a mob of angry llamas aiming their spiteful rage in vain at the sight of a majestuous mechanical flying unAIcorn. At least that's how i picture it. I think it's unfair to the mod your provided. I wanted to post a picture of chad engineer fishing in a pond with a bait that look like a box with "AI" written on it. And the pond be filled with ignorant piranas trying hard to peer pressure themselves into hive mind after seeing your second mod but i couldn't get an AI to properly make visible their feeling of entitlement and not offend any llamas, It's just much less effort to write the bad prompt so here we go, sorry for digression.

I don't think i'm going to use this mod to solve any actual problem with combinators, i consider myself a veteran, and would use it just to have fun with what i expect would be broken contraption, to toy with AI and measure their progress. However what i would find interesting is to see if the AI is capable of doing things like calculating a square root like this https://forums.factorio.com/viewtopic.php?t=40747 which i often time see people ask about on forums or discords, or a GCD like https://forums.factorio.com/viewtopic.php?t=113159 which took me quite some time and would be an example of valuable result to me justifying learning how to use your mod beyond curiosity of, "where does it fail" "why" and "can i guide it enough to make it succeed", "what would it take ?"

I think it would fail because those are "too difficult" for LLMs nowadays in december 2025, maybe in a couple years, that's my thought before any actual attempt, just my anticipation, to make sure i don't commit confirmation biais. It could also very well be that i'm too tech illetrate to write proper test cases or divide the problem in smaller task and that would be unconclusive on the AI capabilities but who would blame their own failure in this situation when it's so easy to blame the AI.

Now you seem familiar with the mod, on the videos ( which i had to download because for some reason they don't display embedded ) you seem to have written many test cases, i think it could be interesting that you show the "limit" of the AI, one example of things that it cannot do, even when you write the tests properly like you would for sure compared to me. Show us the limits, how it look when it doesn't work to make it easier for user to recognize when it's not a mistake but just the limitations of AIs with "complex example". Maybe even a way to workaround a few rookie msitake, what is a "not enough specific prompt" and how you correct it. How you wrote the test cases differently, like low level, detailed, step by step explanation, to me otherwise it feel much confortable to stick with what's familiar. ( for it to be a tool and not a toy ).

Of course there are different models one can use which could lead to varying quality in the output, but you listed a couple, may as well pick one as an example, ( if possible one that isn't "paid" ).

Besides it's friday, and i would enjoy reading more about your mod as if it was an FFF, not necessarily use it myself (sorry about that) but learn more about the workflow when 'someone' do so, sharing your experience as user too, what did you learn, "honestly". It's new and thought provoking i find, for the better or the worse is up to you, i hope for the better despite the biterness of a few, i think it's fair to assume we can all learn from your experiments even when not participating directly.

It certainly seems like mob of angry llamas at times (even more funny that llama is an LLM as well). People are of course free to like and dislike whatever they want and I'm fine with that. I was just hoping the criticism was a little bit more nuanced than "AI bad".

I can see that as a veteran you would like solve some of the challenges yourself. However, also as a veteran I can see it interesting sometimes, because calculating the GCD for instance isn't something that is solving some Factorio problems by itself so you can combine those parts which are very hard to realize in circuit logic with the AI combinator. The tasks you mentioned definitely can be solved these days already by LLMs, here's an example with the GCD:
https://pub-9645bce205344b639c5666a0559b5528.r2.dev/gcd.png

Square root would be trivial because it has access to a square root function.

Agreed, there would be a lot more to show, but it was already like a 2 minute video, going in detail is going to be very lengthy. :D Should probably have shown a more complex example to highlight the capabilities. I'll consider making some more video content, to dive deeper into some of aspects, but making videos isn't my strong suit so it takes me a lot of time. For the time being the in-mod help section offers some advice for common pit-falls one might encounter.

The thing is because it uses API access, all online providers want some form of (albeit small) payment for it. I have since added support Ollama, so users can use their own machine to run LLMs for free, but that requires some knowledge to set up and ideally a strong computer.

Yeah I could think about writing some of the inner workings, which are considerably more complex than "AI does it all" as some may think. Not quite sure what the best communication channel for that would be though.

Pretty sure it's only the LLM that's spelled llama , or maybe something in spanish, i don't know i do typos often :) in my langage the metaphora goes with toads spitting on doves, but in english toad is also that cute nintendo mushroom and the lama is kinda like a natural spitting sheep so i thought it was a more fitting comparaison for the kind of behavior albeit maybe unfair to lamas, maybe hyenas would be a more comon representation, sorry again for the digression, i have nothing against lamas.

And again for some reason i have hard time accessing the image, i suspect it's because "pub" means "ads" in my langage and i have adblockers running everywhere, i can see it,but not in browser, and anyway, it doesn't really show how it look in game, nor the 2 videos on the modpage, because the UI is always open. Makes it hard to imagine how it looks when calculating the GCD or the square root, but it helps understanding how the GCD is "solved". I can see how one can still fail to write the proper test cases like the mistakes i commited when doing it without AI and someone on the forum pointed out hehehe. I think it's cool because when you have no idea what AI ares in practice you can easily think the AI does everything and jeopardize the need for any skill after you click a button, but the fact that you have the test cases written in formal logic makes it closer to a wolfram alpha than a LLM/chat-bot to me, from a user perspective. And the workflow with test cases sounds oddly reminiscent of the video showing how the factorio game is developped, those would be reasons for me to try and learn how to make it work more than the need to solve an actual problem with combinators x).

Similarly as I agree that the GCD isn't solving some factorio problem, i tried to use it to make music, but other time i just like "solving puzzle" it's pleasant for me to click the wire, test, copy paste things around , add remove, and do the research on the internet, learn how it was discovered or invented, who uses it and why, try to do the same in factorio, to learn new stuff in general because the factorio game offers problems to which i'm already familiar with several solutions for playing a lot over long time. Your mod i see as something interesting like this, a puzzle to try and make it work , learn how to use, without necessarily a problem to solve with it. The mod is the puzzle x). the pay-thing was a big no-no for me to use but that wouldn't prevent me from trying to understand what it is, how it works and so on ^^, i have to check if i can run the Ollama in reasonnable conditions, on local that sound a big big + to me in usability i prefer running older or less good model but on my laptop so i can blame "AI bad i knewit " like a llama if it fails.

Sorry if you thought "now the mod is done it is over" and i'm there asking questions instead of using it but i have to agree with you on the "there would be a lot to show", i know videos takes a lot of time to do ( and computer ressources like AI too ! ) , i'd be happy to old school "read" some more , i know the official factorio forum has a part where people can present their mods where you can upload a few pic, small clip, or links to more infos, I think for this one more than for the other it's justified x),

Also sorry if this is offending, it's long message already, so to say it straight, i'm less interested in the presentation of the "AI product for customer", i'm already convinced , more "101 instructions on how to use for absolute idiot"( good to hear there are in games tips may have missed that ) and/or "back story , how ?" i don't want to be clicking some random mod on the portal, i wanna know more, like is it the new RCON thingy that allow this to be a thing ? is it safe ? like if you say it's a 2 hours joke project for you as a student who has plenty other projects or a team of veteran engineer worked on it a year i'd equally want to try anyway but i'd be proceeding diferently. I've been playing this game since 10 years, i didn't expected this to happen at first but there were always talk about "AI in the future", and now it's here, so i wanna know more, the devil is in the detail some people say, and want to make an informed choice, to not act like a llama on a superficial judgment ! x)