Viewing profile — SkitterKherpi

They'll probably be back later. It's very possible they are "saving up" for a much bigger run.

$5/$25 for Opus 4.8 vs $3/$15 doesnt seem cheaper enough to be too worth it. It depends how much better it is than e.g. Mimo, but I imagine Mimo and co to be too cost efficient in …

I assume you have the dgx spark? At this point I am not 100% on the difference other than Linux and Windows. The RTX spark should come around Q4, unless I am mistaken.

You can run 8bit 27B models at 24GB, it's definitely enough for the model size.

10k is rather a lot yes. For LLMs you can use a lot of tokens with 10k with less hassle without the machine (and also it's not like electricity is free), but for some other things …

I am considering getting something like NVIDIA's RTX Spark when it comes out, though even that will be limited to 128GB.

27-30B in general seems to be the level where you actually start having decent models. I just wish consumer hardware hadn't stagnated so much that we can't easily go higher than th…

Gemni is really not that far behind, 3.1Pro was the best model or close to it when it came out, 3.5Pro likely wont be but whenever they get to 4 that could easily be Fable level, j…

So them banning Fable for only non-Americans is what we non-Americans should expect to be the norm going forward? Way to build even more resentment abroad. I'm very pro-west, but a…

I like the idea, but I am worried that it's yet another step on the road towards personal multi-purpose tower PCs built part by part no longer being a thing.

It has so far been the kind of thing that always feels like the next version of the local models would be the one that is just good enough.

It is impressive how well they've scheduled all their releases, posts, and other news to dominate the tech news cycle almost every day in this pre-IPO phase.

I can see it but even if I do that for something like tests I'd still eat the time cost of the normal Gemma for 10% extra performance. And further, if you switch between the fast a…

It is cool but local models while okay already feel noticeably worse than even the cheapest APIs so I can't see myself sacrificing even a little bit of their quality for speed. I'm…

There's a lot more examples of grep usage than Visual code search in the training set.

It also tried to force usage the paid Claude API instead of claude code usage just because there's a mention of another provider we might want to plug in (which hasnt even happened…

Since the EU has been under investing in nuclear, solar + batteries investments is the only way forward. While the EU is making good progress, I still think it should invest a lot …

This reminds me of GANs and generating directly from a latent vector, as well as exploring them. I wonder how many of the original tricks used with GANs (like extrapolation stuff, …

Perhaps this mainly tested a subset of them but context files as long term memory clearly help in general. The models make the same type of mistakes without them, which are often t…

There are definitely things Deepseek beats GPT on - cost effectiveness, lack of reluctance on some tasks, but from using most models I doubt it actually outperforms in a useful way…

I've never associated it with speed either due to UI. It's not slow, and it does feel snappy compared to some of the older alternatives but it's never felt particularly fast. Grant…