byzantinegene

on the contrary, AI seems to do better at high level code generation

if your boss is prompting the AI, that's what the claude output is going to sound like, if not worse.

so you're saying that junior software engineers write better code now? I find that hard to believe.

I don't see why that is a concern when decomps and recomps are already blooming, they can copy your app down to the atom.

hooked to ai coding, but not tied to any particular model. if they decide to bump prices up, I can easily switch to a cheaper chinese model on openrouter.

we're not talking about most humans here, we're talking about the capabilites of the universe of humans.

AI is very task-driven, it would do whatever is necessary to complete the task, no more. All these evidences of AI going rogue is just them doing whatever they can to achieve the task.

the people are good at using AI are the same people who are good at developing software systems. traits like curiosity, knowledge-hungry and desire for continuous learning are not common in the population, so eliminating the coding barrier doesn't magically increase the number of people who can become capable developers.

I think it's twofold, there are less people willing to learn the trade when most of it can and has been replaced by llms.

svelte is like go, and react is like javascript. alot easier to write non-performant javascript then non-performant go.

90% of software engineers are not, so they don't need a premium model.

one moment it's superintelligence and one moment it's a child?

it's the AI that's 10x not the devs, why should the devs be getting a raise?

yes, because you're billing me based on hours.

why would developers be paid more if it's the capital investment (i.e. the machines) that's leading to more productivity?

presumably...

especially if a good chunk of them are out of work mathematicians and physicists due to ai

you punish them hard enough their investors will have cold feet to keep the business going.

i'm on chatgpt plus and astra high uses up my 5hr limit in a min.

I find codex to overengineer for routine tasks compared to claude. I use codex mainly for infra tasks.

they could potentially lose their license, I think that's worth caring about.

amicable divorces are like 5% of all divorces, so there's not much to be won there

they would if they could, their investors are seething already.

22x more to be exact

i think the main consensus here is that the actual performance is not indicative of the benchmark performance (which supposedly outperforms the previous iterations)

i'm not sure why there's a need for comparison here. both are not saints and you shouldn't trust any of them anyway

I think the (non-definitive) evidence is that the chinese models are always just slightly behind the publically released US models, and never ahead.

China is still quite far from monopolizing hardware production

most of the bubble right now is fueled from circular financing of Anthropic and OpenAI

100b sounds implausible if you consider that anthropic's revenue is only expected to reach 100b this year, 100m is more likely.