Making games in Go: 3 months without LLMs vs. 3 days with LLMs

The LLM started with a three month headstart, both in terms of code, using the previous game as a template, and more importantly, all of the learnings and mistakes you made in the hand-coded pass.

But they're right and you're not addressing anything

I don’t get these arguments.

Just because LLM learned the entire internet does not mean it can’t 10x a developer.

Were all standing on shoulders of giants - LLM or no LLM. LLM gives you extra pair of ladders. That are wobbly for sure. But ladders still.

Using LLM invalidates nothing and the 10x stories are totally plausible.

The plagiarization argument is to me moot - most of us are plagiarizing previous work, weather we know it or not.

” but it only did so by stochastically parroting hallucinated slop”

I’m not sure what the argument here is. That you should only use the most primitive tools available?

”LLMs suck” arguments sound exactly like ”IDE:s suck! True engineers writes everything in notepad.exe”. Or ”debuggers suck! Use only printfs!”

People have different ways of cognition and different strategies of problem solving and tool use. The fact you can’t understand other persons tool use is not an argument against those tools. It simply points out the tools may not be usefull to the critics mode of cognition or problem solving. Not that the tools suck.

I was being ironic! I thought the chain of greatest hits in that phrase would give it away, if not the following words.

That wasn't the argument at all.

If I wrote a card game in 3 months, then cloned the repository and used that as a base to implement a different card game, it would be significantly faster with or without LLMs.

The argument is that it's not really an apples-to-apples comparison. "I wrote a game in 3 months and then used an LLM in 3 days to fork it and make some changes" is not the same thing as "I spent 3 months doing something that I could do in 3 days with LLMs".

I believe LLM-driven development is powerful (though I hate seeing "10x" everywhere, and have hated it long before LLM coding was big, when it was talking about supposedly superhuman coders), but this is not a good comparison, and the argument is sound.

Not only that but it was also their first game, meaning they faced a ton of unknown unknowns which are no longer there. If they were starting to program a card game today without LLMs they would still be able to apply all the knowledge and insights they gained from the previous experience; it would take significantly less than three months.

They're being sarcastic, they're parodying common objections people have about AI, it's not about the technology at all but about their feeling of obsolescence which impacts their pride.

Yeah, I figured this was clickbait but my jaw still dropped a bit when I saw this:

  I cloned the backend for Truco and gave Claude a long prompt explaining the rules of Escoba and asking it to refactor the code to implement it.

How long would it take the human dev to refactor the code themselves? I think it's plausible that it would be longer than 3 days, but maybe not!

I don't know I feel like rewriting a backend for one card game into a backend for another wouldn't be that difficult, especially for the original dev. Once you've worked out how to represent cards and code the rules you're basically there for any card game.

Also, a refactor is by definition rewriting code without changing the behaviour. Worth knowing the difference.

Exactly true. I've had the experience of implementing a programming language in two different languages. The first time it took half a year. The second time it was done in a month, because it was really only a matter of translating all of the code from one c-style imperative language to another, and swapping out the important library code and function calls.

Maybe AI can take that process from a month down to 3 days, and that's great and all, a 10x improvement! But you don't go from 6 months to 3 days without doing the 6 months first.

As an LLM hater, I have to say, this is exactly the use case I want code generation for. If I need to figure out the problem as I develop, which is the case for new code, the model can kindly get out of my way. But if I have already written a bunch of code and I can explain the problem with the understanding that I've gained from my implementation and have the bot redo the grunt work? fine with me..

You don’t even need to clone it. I’ve re-created months long projects from scratch and next time around it takes me around 1/3 of the time of the last, ballpark.

>As an LLM hater

I thought this was the start of a joke or something, I guess if you use LLMs you are a "LLM lover" then.