lisplist

Intel GPUs are pretty worthless right now for inference, let alone training. Which isn't to say that there aren't people getting value out of them, but from a dollars/wattage/sanity to performance, there's no point to buying anything but Nvidia for most people and businesses. I don't want this to be the case, but that's just how it is currently.

indeed

I guess GPUs are interesting because unlike houses, we've suddenly found a massive amount of new utility for these things. If my house suddenly started producing its own furniture, it might be a bit more valuable. Which isn't to say people haven't found new things to do with their houses - COVID made having a home office a lot more valuable, for example, but nothing quite as dramatic as GPUs suddenly able to produce their own code.

There are lots of governments around the world - some big, some small. Lots of discussions around digital sovereignty are framed in this manner - a model for Brazil, a model for Japan, etc. There certainly could be coalitions of governments or governments could strike out on their own. Whatever floats your boat.

The problem is the massive amount of money, data, and talent to produce anything remotely competitive. Ideas have been floated about distributed training, but training requires insanely fast interconnect between nodes making it impossible with the current methodologies we have. Now if you mean funded at a national level, that seems more doable so long as the training runs are centralized in big DCs.

I’ve very mixed feelings on Mimo. The new models do surprisingly well on some tasks, but they don’t allow you to set the thinking level and the model often times spends next to zero time thinking about a problem. If you coax it just a little bit, you can get it to solve it, but this isn’t an issue with other models. Also it’s incredibly easy to get it to get caught in a thinking loop.

Small nitpick, but you can use any model you want with Claude Code and Codex - doesn't have to come from Anthropic or OpenAI

Yeah, AI has really taken the fun out of niche languages for me. Not that Ruby is all that niche, but just learning different programming languages feels a bit pointless now since those are abstractions we built for people, not for bots. Sure, nothing is stopping me from writing code by hand for hobby project, but a big reason why I enjoyed learning new languages so much was precisely because of the productivity boost I might have gotten of that new language. It felt good to become an expert in domains few even knew about. Now, it doesn't matter if your app is written in Python, Ruby, Elixir, Rust, whatever, the AI can achieve it all the same.

For the handful of models you can actually use through Cerebras, they are incredibly fast. The problem is that they aggressively rate limit their APIs to the point where if you just run a single coding agent, your coding session will constantly stall out just due to how fast its able to hit those rate limits.

If you made the mistake of buying their stock the past few weeks, you might wish you hadn't (or maybe it's an opportunity to buy low yada yada)

Perhaps it is different dates for different people then

Have they? AI agents only really got usable around last fall. LLMs were pretty useful starting basically at the beginning of 2025, but not in the fully autonomous way that they are now.

Tbf you could get them as low as $3500 a few weeks ago

My understanding is if something is entirely absent from the dataset, good luck trying to teach a model that through post training

I personally rarely touch an IDE anymore. I talk to lots of people with the same experience. This is anecdotal, sure, but if you ask any software developer that isn't staunchly anti-ai or just living with their head in the sand, you'll probably get told something similar.

H3, the video model. Hy3 was okay

MiniMax models have been pretty disappointing. M2.5 was good, M2.7 was meh and came with a terrible license, 3 meh. Maybe this will be better, but I have my doubts. When I say good vs meh vs bad, this is all relative of course. If M3.1 came out a few months ago it would be groundbreaking. Things just move too quick.

H3, on the other hand, is phenomenal

Gonna be waiting a while streaming your weights from disk

My interest is mostly getting solar installed at home, something I can feasibly do. I don't think my city allows for residential wind turbine installs.

Just chiming in that the idea is cool, but the speeds are way wrong

If only Nvidia or a US lab could put out an open weights model that isn't a year behind Chinese open weights.

My car had a few dents in it and leaked oil so I decided the best course of action was to set it on fire.

This is a fair point. Maybe models become better than me at predicting what I need.

DeepSeek V4.1 Flash is mixed MXFP4/MXFP8 so all but the INT4 calculation is wrong here and that's still wrong because you can run it on < 400GB VRAM. The n-gram table is MXFP8, but can be offloaded to RAM or disk without too much of a performance hit. Really, you could probably cram it onto < 300GB VRAM if you're willing to apply a small quant to certain parts of the model considering how little VRAM is dedicated to kv cache.

I know my comment is a little nit picky because it's still pretty expensive to run, but it's not quite as bad as this comment makes it out to be. Really, if you're VRAM constrained, take a look at GLM 5.3 Flash or Qwen 3.8 Flash Next before you worry about this model as all three models perform pretty similarly.

True, I was thinking about this rebuttal but decided not to include it in my comment. There's a difference between not choosing to take a refugee into your home vs actively making that refugee's life worse. Similarly, you can't fix factory farming on your own, but you could skip meat once a week to make the problem slightly less bad. There are so many issues though that we all have to pick and choose what's important to us.

My hope is that AI, while probably causing great societal turmoil in the short term, leads to such abundance that a) everyone can live a dignified existence, and b) we'll have such great alternatives to animal products that nobody will chose to consume animals anymore due to its replacement either tasting better, being cheaper, etc.

The cynic in me says we'll all just be rendered useless and disposable by AI, but I'm doing my best to look for silver linings for the sake of my own mental health.

I'm not a very good mathematician, but I do know a fair bit about software engineering, and with AI I've been busier than ever. I probably wouldn't be so busy if AI was better at anticipating what I actually wanted rather than making guesses no human would ever make.

This is just a short term problem though. Eventually AI will get pretty good at figuring out exactly I want and it will build that from the start. The requirement of me reviewing the AI output only lasts as long as models stay bad at anticipating my needs, which I don't think will take too much longer.

Not to derail, but the optimist in me thinks if we suddenly gained the ability to converse with livestock, we'd stop eating so much of them since they could tell us how much they suffered.

The cynic in me says it wouldn't change a thing as plenty of people know the horrors factory farmed animals face and still continue to consume them anyways.

Hopefully GPT 8 will treat as a bit better than we treat the cows.

I just mean if AAA games become inaccessible, I imagine there will be large community efforts on creating alternatives to that.

True, but if we (eventually) have models which can produce high quality games, who cares?