We have updated the web version of AITuber OnAir and added the following new LLM models.
===Z ai=== ・glm-5.3 ・glm-5.3-flash（Vision対応） ===DeepSeek=== ・deepseek-v4-flash-vision-exp（Vision対応） ===OpenRouter=== ・anthropic/claude-sonnet-5 ・anthropic/claude-opus-4.8 ・z-ai/glm-5.3
This is the conspiracy nobody wants to say out loud
Moon dev: "didn't the hedge fund build DeepSeek, and if they did, what were they actually building it for" "what if they made it bad on purpose and the whole thing is a setup"
New Delphi Podcast with accelxr, author of our latest report on China’s AI labs
He went into the research thinking China was mostly distilling US models and benchmark-maxing. He came out with a very different view We cover: – Why DeepSeek is doing real research, not just
Over the weekend, video understanding was added to deepseek in brazosd.
It can find good movements from long training videos, then disassemble and edit the movements, and then review and improve the movements. Sports action analysis should require a relatively high FPS, and it can do this, which is really powerful. It’s shocking to read a 1-minute tennis training video, and the token is less than 10 cents.
You have reached the end of the archive
All of deepseek