DeepSeek V4

If those docs were written by Deepseek, it’s also a pretty positive review of the model.

Oh that's quite interesting and hasn't been my experience with regular backend code specifically with respect to tool calling. However that could be because the tool calling format in vllm for Deepseek v4 was broken until a few days ago and that's how I'm running it. I've been hearing amazing things about Flash, I should give it a try.

We go through this with every startup cycle. Startups are not expected to be profitable because they’re spending so much money on growth and R&D. The concept of running a business in an intentionally unprofitable state is confusing to those who don’t understand startup funding. The weird thing is that so many people believe that inference is unprofitable. There are large open weights models that companies run at a profit while charging far less than what OpenAI and Anthropic charge. Deepseek V4 just made their 75% off deal permanent and it was already very cheap. Yes, you have to consider costs of training the models, but as usage grows it’s going to become a smaller and smaller part of the business. I think we will see some data center businesses and AI companies blow up, but I think the people expecting the entire AI scene to blow up because prices quadruple are going to be disappointed.

Actually the fact the inference of a SOTA model is completely Nvidia-free is the biggest attack to Nvidia every carried so far. Even American frontier AI labs may start to buy Chinese hardware if they need to continue the AI race, they can't keep paying so much money for the GPUs, especially once Huawei training versions of their GPUs will ship.