roVinchi

3 ms·

1. 1.▲
   by roVinchi 1y ago

   Can users poison the future training dataset by ending every chat with a dissatisfied feedback even though the chat helped them? Or be even more malicious and steer the chat into destructive behavior and then give very positive feedback?
2. 2.▲
   Possible Open Source CUDA Alternative by Intel, Qualcomm, Samsung, Google?(github.com)

   1 points by roVinchi 3y ago|0 comments
3. 3.▲
   Equivalence of Associative memory and Attention in Transformer models(brandstetter-johannes.github.io)

   2 points by roVinchi 6y ago|0 comments