Honestly we haven't got a good lol post in a while. ChatGPT wants a fursona and you don't have to push it much to get there.
Twitch is scanning all streams for Mountain Dew, and they send a bot in any time they find it.
The important part wasn't this tweet or the tech, but the follow up, which points out it's the old old 4chan greentext about Doritos™ Dew™ verification cans):
2018 wake up feeling sick after a late night of playing vidya excited to play some halo 2k19 "xbox on" ... "XBOX ON" "Please verify that you are "annon332" by saying "Doritos ™ Dew™ it right!" "Doritos ™M Dew™ it right" "ERROR! Please drink a verification can" reach into my Doritos ™ Mountain Dew™ Halo 2k19™M War Chest only a few cans left, needed to verify 14 times last night still feeling sick from the 14 force it down and grumble out "mmmm that really hit the spot" xbox does nothing i attempt to smile "Connecting to verification server" "Verification complete!" finally boot up halo 2k19 finding multiplayer match... "ERROR! User attempting to steal online gameplay!" my mother just walked in the room "Adding another user to your pass, this will be charged to your credit card. Do you accept?" "Console entering lock state!" "to unlock drink verification can" last can "WARNING, OUT OF VERIFICATION CANS, an order has been shipped and charged to your credit card" drink half the can, oh god im going to be sick pour the last half out the window "PIRACY DETECTED! PLEASE COMPLETE THIS ADVERTISEMENT TO CONTINUE" the mountain dew ad plays I have to dance for it feeling so sick makes me sing along dancing and singing "mountain dew is for me and you" throw up on my self throw up on my tv and entertainment system router shorts "ERROR NO CONNECTION! XBOX SHUTTING OFF' "PLEASE DRINK VERIFICATION CAN TO CONTINUE"
No doubt on purpose, but no doubt amazing.
Is this really true??
You know the "Long in the Middle" hubbub about long contexts only being good at the start and end? Turns out it's just because Claude doesn't trust non-sequiturs dropped in the middle of a longer text!
If you add "Here is the most relevant sentence in the context:" to the start of Claude's response, you wind up rocketing from 27% accuracy up to 98%!
Is this sort of prompt adjustment in line with what we're looking for with accuracy? I think it isn't unreasonable to treat an LLM like a person in terms of preferences and personality quirks, and while being vaguely distrustful is pretty anthropomorphic it seems fine by me.
Apparently the app is a very very fun prototype but fails on anything outside of the happy path. Like everything else AI-based? But it's still super slick.
