Raw LLM Responses

I didn't think i'd be listening to you talk to Tony Ord about this, and within the first 10 minutes have the entire basis of a point made on just, a complete lack of fact, but hey ho: AlphaGo was trained on human games up until the point it beat Lee Sedol, after that, and leading up to the game against the world champion (the chinese game) it was then changed to a model where it played itself. And even then, it was still initially trained on human games (they started with amateur games). The argument made that AlphaGo isnt representative, becuase it's level was achieved without human experiences being invovled, is objectivally false, becuase it beat Lee Sedol in a way that people went 'thats one of the most brilliant creative moves we've ever seen, and no human would come up with it', from training on human data. I don't think this version of ai being worked on will end the world or something, but I find it very very baffling that two people who are so dedicated to ai commentary, are both completely oblivious to how deepmind operated. Even alphastar had a period where it played against real people on the starcraft ladder to assist in training it. ANd as for the comparions about a lack of ability to check or test stuff, the only real way that was truly done for alphastar and the dota version (I forget the name of that one), was by playing pro's and looking at what/why it did what it did. Anyway, kind of dissapointed. This isn't even a big complicated fact, it's in both documentaries on deepmind. I thought this might be an interesting discussion, but like, come on lol