I think the root of the problem (lack of access to OpenAI for some people, or a lot of people if you are talking about multimodal or 32k context) is that the op

I think the root of the problem (lack of access to OpenAI for some people, or a lot of people if you are talking about multimodal or 32k context) is that the open models aren't quite competitive yet. So we are relying on this one company with superior models. I hope that efforts like OpenLLaMA can be successful. Ultimately what will really democratize this is new hardware paradigms that are variations on compute-in-memory and vastly more efficient. People that showed up a little bit later might be kind of screwed now though with a long wait-list. Not sure. From my perspective OpenAI has done a reasonable job at distributing access to a limited resource. But they definitely do hold some things back like the 32k and multimodal for privileged companies. That part doesn't seem particularly fair but also it seems that there just isn't capacity to do otherwise. I think the whole landscape opens up dramatically within a couple of years with much more efficient hardware for GPT and much better open models.