It's like Microsoft banning Vim users that use Azure

They didn't ban people from using Claude, though. They banned them from their flat-fee subscription and required that you pay per token.

It's still questionable but I don't think it's in the same ballpark as what you describe.

I don't think it's in the same ballpark at all. I checked the `/usage` in my session which uses a Max x5 plan. One day I had used $400 of tokens and 20% of my Fable allocation. Anthropic is effectively giving us more tokens per $ on the monthly plans but it comes at the cost of Anthropic being the prompt-writers and managers of the agents pretty much entirely. I don't think this is a bad deal.

Whether or not it's a bad deal depends on what you're comparing it to. Compared to API pricing, of course a subscription through CC is a good deal. But when OAI offers their super-subsidized plan and allows you to use your own harness which is 75% more token-efficient, then the CC deal starts looking like a bad one in comparison.

It’s really not. Vim isn’t instrumental to Azure usage.

CC isnt instrumental to use Anthropic LLMs. Yet here we are.

> If there's such a dramatic hole in demand, who are NVIDIA selling their compute to?

NVIDIA itself is also training foundation models (and open-sourcing them). If there is excess compute available, NVIDIA can increase the scale of such models.

What is stopping us from all using IRC again ?

does pytorch count

PyTorch had the "first thing that didn't suck" advantage and now it has a completely dominant marketshare that prevents better alternatives from emerging. Where it sucks (e.g. on macOS) there are popular alternatives. But it's hard to be enthusiastic about a DL framework in 2025 which does not have native high performance quantization support, for example. Or one where FSDP is crudely bolted onto the side. They say "usability above all else", but I consider such things to be major usability deficiencies, which need to be addressed. But because PyTorch does not have to fight for marketshare, it'll be years before we see anything usable there.

I wonder how Meta trains its models. On vanilla Pytorch or they actually have some closed tools and frameworks?

DeepSeek V3 and even Janus has all software open sourced and R1 should be fully open sourced as well soon. More importantly, they explicitly spelled out their methodology in a published paper for DeepSeek R1. Implementation is not as important imo but we'll get that soon as well

https://github.com/deepseek-ai

These types of "repositories" should contain some kind of flag/indication that it contains no source code, similar to when a repo is archived

really? it takes like 1 second looking at the file structure to see what it is, maybe like 2 seconds if you’re hopeful “images” somehow refers to a dockerfile or something