AI progress skepticism is getting louder and more specific

The 'AI is slowing down' thread is generating real debate, not just dismissal. Critics of the piece say the tone undermines the argument, but several commenters acknowledge the underlying claims have merit. The key ask in the thread: show us the OpenAI and Anthropic S-1s. That's the only way to actually know whether the economics and progress are as strong as the hype suggests.

Separately, the xAI thread is more pointed. The argument there is that xAI looks more like a datacenter REIT than a frontier AI lab, that Grok is not competitive at the frontier, and that the company may have pivoted toward infrastructure revenue to cover costs rather than because of a coherent product strategy.

Then the MiMo-v2.5-Pro-UltraSpeed thread cuts the other way, showing a 1T parameter model running at 1000 tokens per second. Commenters are genuinely excited, arguing that speed at this scale unlocks entirely new workflow categories. So the picture is complicated: frontier progress from the American labs may be slowing or at least getting more expensive per unit of improvement, while Chinese open models are pushing hard on speed and accessibility.