Tom Vondriska

America doesn't exist on the internet thomasvondriska.substack.com…

file this with: things self-evident that someone felt the need to publish on

An enjoyable read. My own view is that AI would provide a far superior review to human beings. In fact, I can think of no scenario in which a human’s review would be better. The problem of ‘AI hallucinations’ is a straw man: human hallucinations are far worse, oftentimes from ‘experts’ with the subject matter. AI’s appraisal is corrected by improved data. That of many humans never can be. More broadly, on current trajectory, there will be no reason to publish original research papers in a couple years, maybe sooner (even now there should never be another review article written by humans—thank god). The only way humans retain competitive advantage over AI is if we choose to heavily regulate it, like nuclear technology, and direct it towards eliminating the waste associated with research and medical care (which the authors correctly lament), rather than become another source increasing it. This seems to me an unlikely outcome, because of a collective action problem: right now, everyone is trying to get ahead with AI while still competing against other humans. Humans stand no chance in terms of novel discoveries, creative output and brute knowledge when competing directly against a super intelligent AI. Scientists will become like those who restore old cars or make furniture in their garages: an antiquated and inefficient, Luddite pursuit. Unnecessary. This transition will be painful for those of us entrenched in doing things the way we were taught, but afterwards who knows? One thing is certain: we will live long enough to see it happen.