@kevindavidhayes@karpathy @a_karvonen For sure, it’s no Move 37 or anything. But it’s far beyond what anyone five years ago would have considered a hyperparameter search
@kevindavidhayes@karpathy @a_karvonen Only if you knew about it and had already included the option in your code. If you didn’t, what pre-agentic hyperparameter tuning algorithm would have been able to come up with it for you?
@arjie@paulg That particular suggestion doesn’t work because it’s already the name of a different warbler species. Sort of surprising that this field guide author or the article editor didn’t double check that?
@ZachWeiner Also elsewhere in the same article: "Final /n/, /ŋ/ may be pronounced without complete oral closure, resulting in a syllable that in fact ends with a long nasalized vowel"
@ZachWeiner See the third bullet here: https://t.co/FoIaEdzTmB. Not close to a fluent speaker but I've definitely noticed what you're describing before (along with what sounds like a bit of vowel nasalization as noted in wikipedia, so I don't think it's creating homophones)
@Vidor1@SexyJedi@mattyglesias They publish thousands of predictions so you could: if you say, sampled 100 predictions around 80% and only 65 happen, that would show Silver to be “wrong”. But in fact 538 does such assessments and turns out to be pretty impressively calibrated: https://t.co/WwouvxzGaR
@gaddmatt@its_adamneely Could it be the other way around? I think a bunch of the Children's Songs borrow from earlier compositions - I know #8 (probably my favourite one) is the first part of The Sky (My Spanish Heart).