(the 海娘 at home:)
(the 海娘 at home:)
how do you do these?
how do you do these?
there are some safeguards, but
• open-weight models accept text and images and their refusal behavior can be edited out, and
• even o3 is still happy to doxx cat pics (see the example thread linked above)
there are some safeguards, but
• open-weight models accept text and images and their refusal behavior can be edited out, and
• even o3 is still happy to doxx cat pics (see the example thread linked above)
if u post a TIGHTLY CROPPED photo, the risk calculus is:
• your correct city in top ~10 guesses
• given known city, correct street/intersctn/bldg in top ~10 guesses
if u post a TIGHTLY CROPPED photo, the risk calculus is:
• your correct city in top ~10 guesses
• given known city, correct street/intersctn/bldg in top ~10 guesses
here's a case study retracing a famous doxxing: which hotel is this pic from? (known city)
to solve this, o3 will spend minutes gathering hotel listings, then google each one's pictures and compare.
the correct listing is 10th guess.
here's a case study retracing a famous doxxing: which hotel is this pic from? (known city)
to solve this, o3 will spend minutes gathering hotel listings, then google each one's pictures and compare.
the correct listing is 10th guess.