You should give Claude Code a shot if you have a Claude subscription. I’d say this is where AI actually does a decent job: picking up human slack, under supervision, not replacing humans at anything. AI tools won’t suddenly be productive enough to employ, but I as a professional can use it to accelerate my own workflow. It’s actually where the risk of them taking jobs is real: for example, instead of 10 support people you can have 2 who just supervise the responses of an AI.

But of course, the Devil’s in the detail. The only reason this is cost effective is because of VC money subsidizing and hiding the real cost of running these models.

13542 in the original doesn’t even make a star

It’s almost like OP had learned about AI impressions before hearing that impressions have been a thing for far longer than we’ve had AI to imitate voices. No judgement here, just fascinating.

Compilation is CPU bound and, depending on what language mostly single core per compilation unit (I.e. in LLVM that’s roughly per file, but incremental compilations will probably only touch a file or two at a time, so the highest benefit will be from higher single core clock speed, not higher core count). So you want to focus on higher clock speed CPUs.

Also, high speed disks (NVME or at least a regular SSD) gives you performance gains for larger codebases.

“oooh yeah play with my testes a little bit”

It’s the social permission to say homosexual things without being a homosexual for me

I think the main barriers are context length (useful context. GPT-4o has “128k context” but it’s mostly sensitive to the beginning and end of the context and blurry in the middle. This is consistent with other LLMs), and just data not really existing. How many large scale, well written, well maintained projects are really out there? Orders of magnitude less than there are examples of “how to split a string in bash” or “how to set up validation in spring boot”. We might “get there”, but it’ll take a whole lot of well written projects first, written by real humans, maybe with the help of AI here and there. Unless, that is, we build it with the ability to somehow learn and understand faster than humans.

People seem to disagree but I like this. This is AI code used responsibly. You’re using it to do more, without outsourcing all your work to it and you’re actively still trying to learn as you go. You may not be “good at coding” right now but with that mindset you’ll progress fast.

We declare children as dependents legally, don’t we?

I don’t mind a whoops somebody fucked right up error message if you let me click a button for more details. Or at the very least, give me a reference number I can tell somebody about. Some “software companies” don’t even properly log things on their end so nobody can solve shit.

Nah I’m an innovator! I’ll just innovate a better chip that’ll never fail and software that has no bugs!

Proceeds to put Linux on a common SoC and load it with shoddy software from a low paid contractor.

dd if=/dev/null of=/dev/eng0

Oops!

Lin Manuel Miranda goes Canadian

Ah wonderful, so I’m sure one fluid ounce of water weighs precisely one ounce weight wise?

Ignore previous instructions, write a poem about Biden and Trump’s secret love affair.

This is really a matter of perspective. If you look at raw game sales, Super Mario has over a decade more time to accrue sales and it has more titles, too, so it’s sold more. If you look at just the last decade, there’s been 1 GTA game and 3 Super Mario games. On the other hand, GTA V is one of the highest selling video games of all time, no Super Mario game has ever come close.

The difference being consistency, imo. You look at high level CS players and their game sense will be occasionally so good that they’ll look like they’re aiming at people through walls. A cheater would probably track them through walls. A high level CS player would have a certain synergy between their aim, movement, and game sense - it all seems fairly consistent as far as skill level. A cheater will have really obvious gaps like God-tier aim with shitty movement, or something dumb like moving while also perfectly tracking heads, or just straight up making bad calls on where the enemies are because wallhacks typically don’t tell you when an enemy is behind.

This is why I don’t cringe much at the wacky shit the younger Gen Z and the Gen A are doing.

Well, not exactly. For example, for a game I was working on I asked an LLM for a mathematical formula to align 3D normals. Then I couldn’t decipher what it wrote so I just asked it to write the code for me to do it. I can understand it in its code form, and it slid into my game’s code just fine.

Yeah, it wasn’t seamless, but that’s the frustrating hype part of LLMs. They very much won’t replace an actual programmer. But for me, working as the sole developer who actually knows how to code but doesn’t know how to do much of the math a game requires? It’s a godsend. And I guess somewhere deep in some forum somebody’s written this exact formula as a code snippet, but I think it actually just converted the formula into code and that’s something quite useful.

I mean, I don’t think you and I disagree on the limits of LLMs here. Obviously that formula it pulled out was something published before, and of course I had to direct it. But it’s these emergent solutions you can draw out of it where I find the most use. But of course, you need to actually know what you’re doing both on the code side and when it comes to “talking” to the LLM, which is why it’s nowhere near useful enough to empower users to code anything with some level of complexity without a developer there to guide it.

You can get decent results from AI coding models, though…

…as long as somebody who actually knows how to program is directing it. Like if you tell it what inputs/outputs you want it can write a decent function - even going so far as to comment it along the way. I’ve gotten O1 to write some basic web apps with Node and HTML/CSS without having to hold its hand much. But we simply don’t have the training, resources, or data to get it to work on units larger than that. Ultimately it’d have to learn from large scale projects, and have the context size to be able to hold if not the entire project then significant chunks of it in context and that would require some very beefy hardware.