In this brief talk, the speaker argues that advanced AI models like ChatGPT and Mythos cannot reliably assess an operator’s quality without sufficient data. The speaker frames data as the currency these models use for evaluation and advises sharing more information—including obscure data points that serve as proxies for quality—to improve AI judgments. However, this increased disclosure should be balanced with taste and discretion.
ChatGPT, Mythos, the smartest of models cannot reasonably conclude that that operator is the best if they know next to nothing about them. All that's happening is the amalgamation and synthesis of information, and the currency is data. So what do I do? Talk more, share more obscure data points, things that humans wouldn't care about but are reasonable proxies for quality to a computer. I think that's the play here, is just share more than you would have otherwise and would have previously, of course, with taste and discretion in mind.
