If this analysis is correct, that would update me in favour of wanting to invest in smaller, currently under-funded startups that are alignment-aware, over for established players like Alphabet or Microsoft.
Although I’m not completely convinced that this is completely the case- based on the difference in performance between GPT-2 and GPT-3, it seems to me that we are likely in a hardware overhang right now, which means that the question of who is in a dominant position to decide the future of the world via AGI, is largely determined by who is able to first rally enough hardware to scale current techniques to AGI scale (although this isn’t the sole consideration, only a dominant one). That means that even a relatively minor change in funding could provide a company just enough access to hardware to let them develop AGI slightly before their competition does, which could have huge long-term effects depending on what (if anything) that AGI locks in.
Thanks for the answer, Christian
If this analysis is correct, that would update me in favour of wanting to invest in smaller, currently under-funded startups that are alignment-aware, over for established players like Alphabet or Microsoft.
Although I’m not completely convinced that this is completely the case- based on the difference in performance between GPT-2 and GPT-3, it seems to me that we are likely in a hardware overhang right now, which means that the question of who is in a dominant position to decide the future of the world via AGI, is largely determined by who is able to first rally enough hardware to scale current techniques to AGI scale (although this isn’t the sole consideration, only a dominant one). That means that even a relatively minor change in funding could provide a company just enough access to hardware to let them develop AGI slightly before their competition does, which could have huge long-term effects depending on what (if anything) that AGI locks in.