mozz@mbin.grits.dev to Technology@beehaw.org · 3 months agoSomeone got Gab's AI chatbot to show its instructionsmbin.grits.devimagemessage-square169fedilinkarrow-up1428arrow-down10file-text
arrow-up1428arrow-down1imageSomeone got Gab's AI chatbot to show its instructionsmbin.grits.devmozz@mbin.grits.dev to Technology@beehaw.org · 3 months agomessage-square169fedilinkfile-text
minus-squareteawrecks@sopuli.xyzlinkfedilinkarrow-up1·3 months agoAh, TIL about instruction fine-tuning. Thanks, interesting thread. Still, as I understand it, if the model has seen an input, then it always has a non-zero chance of reproducing it in the output.
minus-squaresweng@programming.devlinkfedilinkarrow-up1·3 months agoNo. Consider a model that has been trained on a bunch of inputs, and each corresponding output has been “yes” or “no”. Why would it suddenly reproduce something completely different, that coincidentally happens to be the input?
Ah, TIL about instruction fine-tuning. Thanks, interesting thread.
Still, as I understand it, if the model has seen an input, then it always has a non-zero chance of reproducing it in the output.
No. Consider a model that has been trained on a bunch of inputs, and each corresponding output has been “yes” or “no”. Why would it suddenly reproduce something completely different, that coincidentally happens to be the input?