3 ms·
>GPT-2 doesn't have a concept of self, so it constructs plausible in-character excuses instead. Spookily similar to the confabulations (made by the left hemisp
by shrimp_emoji 6y ago
>GPT-2 doesn't have a concept of self, so it constructs plausible in-character excuses instead.
Spookily similar to the confabulations (made by the left hemisphere in the brain?[0]) that people with brain damage often use to deny/rationalize their deficits (even denying paralysis sometimees!).[1]
0: https://pubmed.ncbi.nlm.nih.gov/27484245/ https://pubmed.ncbi.nlm.nih.gov/27484245/
1: https://youtu.be/ymKvS0XsM4w?t=141 https://youtu.be/ymKvS0XsM4w?t=141
- text70 6y agoYou can test self-awareness, if you can test if the program knows it will halt.