latentsea

Doesn't deepseek make $38 per GPU-hour? As in they can actually serve profitably?

> My local model never says no. I never get banned.

So much this. Plus everything else you mentioned. The freedom and control is real.

They dabble in pulling apart culture.

This got a chuckle out of me.

Slop

This. I don't know why people think the goalposts have moved.

Try it with Qwen2.5 then.

What if we're really just living in a world model? How big of a GPU is required to run the universe I wonder?

Nope. The stuff I need at home is just stuff I need. It just has to work for what I need it for. The development is basically interact with it through the UI and tell the agent what the bugs are until they're ironed out and then I never look at the code. True vibe coding. At work I need to make sure the system properly works and is maintainable and I need to understand it, and I'm sick of reading Claude and processing information from the wrong end in way that's really suboptimal for my own understanding.

I definitely know what you mean about the difference in engineering culture. I'm definitely the kind to get excited by it and try out all manner of experiments and really have my head in the game on the new paradigm. My current team just isn't in that same wavelength. It's a lot more practical focus on shipping, and fair enough. It's a bit of an impedence mismatch for sure. Still, I have fun playing around with stuff in the evenings at home, but now at work I'm just trying to do bog-standard engineering and forget about the AI side of it to the degree that I can.

>it's watching people blindly use AI in a way that you know, being experienced, will end in disaster. And knowing that any argument will be dismissed under the pretense of obsolete thinking.

Experiencing this led me to feel burnout. It's horrible. I now value thinking and the ability to think much more. Not that I didn't before, but I mean I have a whole new appreciation for it and a whole new disdain for Claude slop where someone could have otherwise done a better job.

I figure if the industry is only going to last a few more years (according to some anyway) then why not just spend as long as I can working the way I enjoy, rather than just jumping on the bandwagon? Actually, I jumped on the bandwagon and rode it for a year, and decided it kinda sucks, so now I'm doing my best to go back to just coding as much by hand in my professional life while just switching to pure vibe coding in my private life.

>This makes no sense. Everybody I know, including me, who hates the big picture of what AI has done loves programming immensely and does it as a hobby as well as a job. The people at my job who are most positive about AI are those who didn't particularly enjoy programming. You're applying Occam's Razor backwards.

I think I know people in both camps who feel both ways. I'd say it's all just noise at this point. What's clear is some people like it (for now) and some people don't (and probably won't ever). I've gone from liking it to not over the course of the last year. I used to love programming/the craft, and then got burnt out of it, so I loved it because it made building things easier again after I stopped caring about that, and then it made work hellish when everyone basically stopped caring about the craft at all.

I think it's more that using the end result of vibe coded stuff is the fun thing. The vibe coding itself is more just a means to an end. I do this for various machine learning things. I have no idea how to do shit in that realm, but my agents do, and so I get them to whip me up things I need for labelling datasets and then training little models to help me with things I need etc. They fumble around and produce buggy stuff and I tell them 'no not like that, change the layout to be more like this' etc until I get something that works for the stated purpose and then using it is fun.

That definitely can result in having fun. It's distinctly different from crafting something non-throwaway you actually care about vibe-coded or otherwise.

I just use the strongest model I can run at a non-frustrating speed. That's never frontier performance on local hardware, but most of us running locally accepted that a long time ago.

Currently on R9700 + 64GB of RAM I'm running Qwen3.8-Flash-Next. It's decent.

13 TPS is painfully slow?

20 t/s is about the speed you can read at, so slower than that and it always feels like it's buffering. 35 t/s is where it starts feeling decent and 50 t/s feels nice.

The only marketing strategy left will be that of "authenticity" and "human made".

I'd buy one for the police chases. Hell yes. Not that I'm in the billionaire invite-only inner circle, and not that they'll ever ship it.

We'll be free of cancer, and dead on the inside.

Krea3 open-weights will be a legendary release. For me Krea2 was already one of those unforgettable drops. Same like Minimax H3 or Qwen3.8-27B.

Reminds me of my favorite quote - "Computers do what you tell them to do, not what you want them to do".

"I need you to cleanup the footage"... "Fuck! Fuck! Fuck! No! No! No! No!"

Intent. Ain't it a bitch?

At some point on this ride, if devs are really put out of work and wages are massively depressed (along with the devs), then it might become cheap enough to make work.

I spent time building the factory, and it was fun. Until it came time to need to maintain the factory, at which point it was no longer fun.

I assume frontier labs will start hiring devs specifically for generating training data for this purpose. It's a skill you can get better at. I don't see in the limit why it couldn't be solved, but I agree, for practical reasons it's somewhat unlikely.

It's true though. I've also experienced the difference between a codebase that was an absolute tarpit from the start, and one that obviously focused on quality from the start. They both became jankier over time, but one notably less so. The latter was a bigger, more successful organization. Does that have anything to do with their codebase being cleaner? Maybe. Maybe not. It was certainly easier for me to work in.

Maybe the solution is just to hire someone who likes to fix janky crap full time and have that be their job. Honestly, I'd do it. I hate jank, so I like de-jankifying things. I don't think I'd mind such an arrangement personally.

> If there's one thing I've learnt about criminals (and spies), it's that a large percentage of them are incredibly careless and arrogant, often bordering on idiotic.

What's even more amazing is how long they can go in this fashion without being caught.

I wonder sometimes wonder what the true rate of this is. You'd never know.

d4vid's entire catalogue is a confession.