2 ms·

> Maybe I am generalising, but the most capable models will always be in the hands of the corps that have the money for compute. This is an interesting thought

> Maybe I am generalising, but the most capable models will always be in the hands of the corps that have the money for compute. This is an interesting thought. Why wouldn't you expect the government to have the most capable model? They can print money so they have effectively unlimited funds. Like when we landed on the moon or created the atomic bomb, if the government wants to do something it usually can. They could literally use eminent domain to get the GPUs, or just use their unlimited funds to offer NVIDIA double what anyone else will pay. In fact, I wouldn't be surprised if the DoD or DoE (or the two together) don't already have a more capable model.

skilled 3y ago
One thing I have noticed (it could be only my assumption) is that OpenAI is quite secretive and generally downplay their recent releases, ChatGPT and GPT-4. What I mean to say is, they often talk about “surprised at the popularity”, but I don’t think its that surprising at all, and they clearly knew they have found something that would change the AI landscape. And at the same time, they can do other experiments because they keep weights and other details about the model secret. I doubt they haven’t already found something far more potent than GPT-4, but are simply planning years ahead of how to trickle the downstream of releases so it makes sense for the public eye and their self-interest.