So I came across the latest on Prometheus, Jeff Bezos's new AI company, and it is a noticeably different from everyone chasing the next chatbot.
instead of text or code, Prometheus is aimed at the physical world. the idea is ai that understands real physics and manufacturing well enough to help engineers design and test actual hardware. Bezos calls the goal an "artificial general engineer" and describes it as a very modern version of cad software. He has also been clear it is not a robotics company, which surprised me.
So although the vision is huge, the demo is the thing nobody can point to yet. And I think the reason is that the entire pitch rests on simulating the physical world accurately, which is far harder than generating text merely. A language model that is slightly wrong writes an awkward sentence but an engineering model that is slightly wrong could lead to unimaginable disaster, so the accuracy bar is very unforgiving. Also there is the data problem like text models had the entire internet to train on but high quality engineering data sits inside private companies and cost quite a lotand often comes from physical testing you cannot scrape. That is probably why Prometheus is reportedly trying to buy industrial firms outright, just to own the data pipeline.
so the missing demo makes sense. shortening the design loop does not shorten the parts of the process that are slow on purpose, because being wrong there is dangerous.
I am not predicting it fails. A team this funded, aimed at a real bottleneck, is worth watching. The honest read is that the demo is missing because the hard part has not been solved yet, not because they are hiding it.
[link] [comments]