David Gerard@awful.systemsM to TechTakes@awful.systemsEnglish · 29 days agocome see all the popular super-duper-autocomplete systems failing hard at really simple reasoning questions and babbling nonsense from latent space!arxiv.orgexternal-linkmessage-square12fedilinkarrow-up151arrow-down10
arrow-up151arrow-down1external-linkcome see all the popular super-duper-autocomplete systems failing hard at really simple reasoning questions and babbling nonsense from latent space!arxiv.orgDavid Gerard@awful.systemsM to TechTakes@awful.systemsEnglish · 29 days agomessage-square12fedilink
minus-squaresinedpick@awful.systemslinkfedilinkEnglisharrow-up9·29 days agoThis all but confirms that all those benchmark evals are in the training set right?
minus-squareDavid Gerard@awful.systemsOPMlinkfedilinkEnglisharrow-up13·29 days agoSome forms are - but many are not! The fun stuff is in Appendix 2, the responses.
This all but confirms that all those benchmark evals are in the training set right?
Some forms are - but many are not! The fun stuff is in Appendix 2, the responses.