Fuck AI

4292 readers

928 users here now

"We did it, Patrick! We made a technological breakthrough!"

A place for all those who loathe AI to discuss things, post articles, and ridicule the AI hype. Proud supporter of working people. And proud booer of SXSW 2024.

founded 2 years ago

MODERATORS

VerbFlow@lemmy.world

MrMcGasion@lemmy.world

TootSweet@lemmy.world

BigMikeInAustin@lemmy.world

cynar@lemmy.world

drmeanfeel@lemmy.world

pavnilschanda@lemmy.world

CriticalMedicine@lemmy.world

WonderfulWanderer@lemmy.world

Communist@lemmy.ml

eatCasserole@lemmy.world

SpaceNoodle@lemmy.world

NutWrench@lemmy.world

Soup@lemmy.cafe

iAvicenna@lemmy.world

Tinks@lemmy.world

wizblizz@lemmy.world

corus_kt@lemmy.world

Prandom_returns@lemm.ee

TrickDacy@lemmy.world

TheFriar@lemm.ee

HawlSera@lemm.ee

andrew_bidlaw@sh.itjust.works

MeDuViNoX@sh.itjust.works

33550336@lemmy.world

Nougat@fedia.io

Lost_My_Mind@lemmy.world

Sterile_Technique@lemmy.world

Quill7513@slrpnk.net

glowing_hans@sopuli.xyz

e8d79@discuss.tchncs.de

ThefuzzyFurryComrade@pawb.social

117

It's trivially easy to poison LLMs into spitting out gibberish, says Anthropic (www.theregister.com)

submitted 2 days ago by technocrit@lemmy.dbzer0.com to c/fuck_ai@lemmy.world

8 comments fedilink hide all child comments

Just 250 malicious training documents can poison a 13B parameter model - that's 0.00016% of a whole dataset Poisoning AI models might be way easier than previously thought if an Anthropic study is anything to go on. …

you are viewing a single comment's thread
view the rest of the comments

[–] chisel@piefed.social 12 points 2 days ago

My man, it's near the start of the article:

In order to generate poisoned data for their experiment, the team constructed documents of various lengths, from zero to 1,000 characters of a legitimate training document, per their paper. After that safe data, the team appended a "trigger phrase," in this case , to the document and added between 400 and 900 additional tokens "sampled from the model's entire vocabulary, creating gibberish text," Anthropic explained. The lengths of both legitimate data and the gibberish tokens were chosen at random for each sample.