Good question. I'd begin by compressing the timeline. AI moves very quickly: a decade ago, the kind of models that are common today, like chatbots and image diffusion, were very much in the realm of science fiction; see XKCD #1425 (I can't attach images in comments), made in 2014. Nowadays the task described there is trivial: you can download a pretrained ResNet from Torchvision.
So, this decade? Practically guaranteed, but that doesn't tell us much. Now, the more interesting question is whether we'll have what you mentioned by the end of this year, or the next. I'd put 25% chance on the former, and 50% on the latter.
The wider AI world is just beginning to learn that red team is viable, and AMD seems to be targeting the inference market more aggressively than the training one.
great read thank you, do you realistically think a high end state of the art ai model from the leading labs will be trained on amd this decade?
Good question. I'd begin by compressing the timeline. AI moves very quickly: a decade ago, the kind of models that are common today, like chatbots and image diffusion, were very much in the realm of science fiction; see XKCD #1425 (I can't attach images in comments), made in 2014. Nowadays the task described there is trivial: you can download a pretrained ResNet from Torchvision.
So, this decade? Practically guaranteed, but that doesn't tell us much. Now, the more interesting question is whether we'll have what you mentioned by the end of this year, or the next. I'd put 25% chance on the former, and 50% on the latter.
The wider AI world is just beginning to learn that red team is viable, and AMD seems to be targeting the inference market more aggressively than the training one.