Post

I mean given how many people I have now argued with it about it is evidently NOT, but.

Oh my gosh, yes. I think of it as similar to riding an e-bike? Yes, I am putting less effort into each pedal. But also this means I'm more willing to just bike places and distances that otherwise I would have driven.

Similarly, LLMs reduce the frustration of coding to a point where I no longer want to throw things when dealing with bugs. So I've spent a lot more time this year doing code-related things that I never ever have before, like deploying to Vercel!

In theory yes, responsibly-used LLMs are an excellent resource for to learn and develop your own skills. I'm sure there's a small number of people using them like this. I'd consider myself among them. I got up to speed on Rust way faster with LLM help than the last time I tried and bounced off it.

In practice, no. Very nearly nobody uses them this way, or at least not for very long. It's too easy to surrender agency and leave it all up to the agent or chatbot, not learning or retaining any information for yourself. They're designed and marketed specifically to encourage this.

So I'd strongly recommend that anyone who is trying to learn programming stay clear of LLMs. If they do use LLMs, use them to find and help interpret credible learning resources, or to help understand errors but only as a last resort after trying other things, and never to write code on their behalf

I think people who learn gamedev via LLMs will naturally gravitate towards using more technical/code-oriented tools. LLMs can't read minds, a user who knows what they want will either need to get much better at technical writing (in natural language) or just learn a programming language.

I've found limited use for LLMs in game *design* because code is ultimately a more concise and easy to edit representation of my project's rules than anything I could write in english. They're great for debugging though.

A side note- Board games and TCGs also run on code, it's just code for humans! The role of a game designer has always been about formally defining rules and instructions for a game, but that's never had to require a computer. It's cool we now have tech that can easily bridge that gap.

people who learned everything from books decrying people who learned everything from the internet decrying people who learned everything from llms decrying...

I see from the rest of the thread that the context is video games, and my answer is that there were already much better options for getting started in game design with no background in programming before the LLM boom. There's GameMaker for one, and genre-specific tools like RPG Maker and Inform 7.

Regardless, vibe-coding is a bad way to get into game design. The way to start is still to pick a language/platform and learn it enough to make a nice-feeling version of Pong from scratch. Getting your own Pong to feel right is the best first step to getting more complicated games to feel right.

You cannot seriously believe you will learn more about game design by cloning pong than by taking your own idea for a game, seeing it realized, testing it, and iterating on the design.

The "technical stuff" isn't the roadblock to learning game design, and hasn't been for a long time. Game design, like most disciplines, is something you have to learn from the basics if you actually want to learn it. Vibe-coding a platformer is getting off on exactly the wrong foot.

As a former academic I take some issue with this, since just having the tool do work for you is not learning. It's going to a restaurant and pretending you're a chef. If you use the tool to help understand what's going wrong with what you made, i.e. as a tutor, that's a lot better

So the situation I was really talking about is one where you have to learn a lot of technical skills before you can get to the thing you want to actually learn.

In particular, if you want to make videogames, you generally have to learn a lot about programming before you can really get to the part where you're designing games. My hope is that LLMs will facilitate people being able to get to a place where they can design a game, test it, and iterate on it...

I feel like that's the worst possible example for your point? Lots of people want to cook a nice meal for guests occasionally. That doesn't require a chef's skills. For most people it's okay to have poor knife skills, buy prepared sauces, and cook from a cookbook.

Executive chefs have lots of specialized skills that an ordinary person doesn't need to cook a nice meal: managing a working kitchen, multitasking, sourcing bulk ingredients, market research, controlling shrink, hiring and training, designing a profitable menu...

The thing is, learning IS HARD. Even under the best of circumstances. There's only so many corners you can cut before you just don't understand what's going on.