Jonathan M

[Jonathan M @jonathanmcdill]

@theo What's the new Arch? I need a refuge to flee to.

[Jonathan M @jonathanmcdill]

@itsmoni_02 @Tushar_Jsx The age of AI is like, you have to know what's going on under the hood, but you better not touch any of it, it's inefficient. It can fix itself now. As long as it drives, only thing that matters. But you better know what's under there. No! Keep that hood closed! Ok.... :(

[Jonathan M @jonathanmcdill]

@Tushar_Jsx Agreed. Problem solving with my own brain was the fun part of programming. Now it's "please fix this problem for me," "still broken, try again." Very unsatisfying.

[Jonathan M @jonathanmcdill]

@chimpansky Honestly it sucked. And now AI sucks in a different way. But I preferred just me and the docs tbh. At least everything I did felt like me doing it.

[Jonathan M @jonathanmcdill]

@mreflow Are you saying code was never the bottleneck, or that it isn't now in the age of AI?

[Jonathan M @jonathanmcdill]

@smol_hu That's cool! People have made little demos like this fairly quickly even before AI--not sure AI is gonna be the unlock here to get you to AAA haha. I genuinely like the ambition and how you open-sourced it though.

[Jonathan M @jonathanmcdill]

@om_patel5 How do you know it was vibe-coded?

[Jonathan M @jonathanmcdill]

@valigo Poor guy probably thought it was real he believes so hard

[Jonathan M @jonathanmcdill]

@cmuratori I think the only way around it is that *I* didn’t pirate the inputs. Anthropic/OpenAI did. But that’s just pushing it back a step I suppose.

I really dislike AI and the fuzziness it’s creating for creative work. Like if you vibe code something, is it really “yours” at all?

[Jonathan M @jonathanmcdill]

@mSanterre Cooking is solved

[Jonathan M @jonathanmcdill]

@rms80 People forget you could copy paste manually before AI

[Jonathan M @jonathanmcdill]

@ID_AA_Carmack A model’s competence is interpolative over the region of code-space its training data densely covers, and that region is human authored code. LLMs predict human thought processes. Code that is understandable by humans and efficient for an LLM will always be coupled.

[Jonathan M @jonathanmcdill]

@nicotwtss Why do people use brain when no need use brain.

[Jonathan M @jonathanmcdill]

@_andrewthecoder I dunno if LLMs are getting worse, but we're lowering our bar for what is actually good to be as low as what the latest model can do. "Because it did it fast."

We're impressed by very little when it comes to genuinely pushing the limits of software. 1,000 prototypes isn't it.

[Jonathan M @jonathanmcdill]

@justalexoki There's no point to bikes anymore, we have cars and they can go faster than bikes.

[Jonathan M @jonathanmcdill]

@bentlegen If only it knew to do that on its own.

[Jonathan M @jonathanmcdill]

I'm impressed with it, mostly because I don't fully understand the complexity of what it did (very little game dev experience). How hard is something like this really? Help me cope.

https://t.co/NhazprhDLj

[Jonathan M @jonathanmcdill]

Game devs, help me out. This is my experience with one-shotting a sim city game with different models. Fable is clearly quite a bit ahead.

👇

[Jonathan M @jonathanmcdill]

@bendee983 Also, the more I use AI (don't worry it's a lot), the more AI-ish it looks and feels. It's like a new comic sans. It feels like a template. It's getting better and better at doing stuff. But it has a cheap "feel" to it. Probably cope tho.

[Jonathan M @jonathanmcdill]

@wawasensei Agreed. A manic "you *have* to embrace AI psychosis and tokenmaxx" is not a good mindset either. E.g. if you enjoy writing code, keep writing code. But if you insulate yourself because of fear, that's not good either. Tech is always moving, being open and curious is good.