[HN Gopher] Poisoning Language Models During Instruction Tuning
       ___________________________________________________________________
        
       Poisoning Language Models During Instruction Tuning
        
       Author : hardmaru
       Score  : 73 points
       Date   : 2023-05-03 12:39 UTC (10 hours ago)
        
 (HTM) web link (arxiv.org)
 (TXT) w3m dump (arxiv.org)
        
       | titzer wrote:
       | No one accounted for the Manchurian GPU!
        
       | egberts1 wrote:
       | This is an important whitepaper as it lends credence into
       | poisoning the dynamic malware detection via LLM.
       | 
       | In short, heavily guard your LLM inputs.
        
       | aurizon wrote:
       | This reminds me of the 'poisoned' clown bike, which has a geared
       | stem that inverts steering, move bars in an arc to the right =
       | bike goes left.
       | https://www.youtube.com/watch?v=oI2aMKwXXnE&ab_channel=MikeB...
        
       ___________________________________________________________________
       (page generated 2023-05-03 23:01 UTC)