positron26

18 ms·

1. 241.▲
   by positron26 1y ago

   And then during the next token, all of that bounded depth is thrown away except for the token of output. You're fixating on the pseudo-computation within a single token pass. This is very limited compared to actual hidden state retent
2. 242.▲
   by positron26 1y ago

   An unrolled loop in a feed-forward network is all just that. The computation is DAG.
3. 243.▲
   by positron26 1y ago

   There are no recurrent paths besides tokens. How may I introspect something if it is not an input? I may not.
4. 244.▲
   by positron26 1y ago

   > because it's perceived as good marketing We are making user interfaces. Good user interfaces are intuitive and purport to be things that users are familiar with, such as people. Any alternative explanation of such a versatile in
5. 245.▲
   by positron26 1y ago

   > Is it too anthropomorphic to say that this is a lie? Yes. Current LLMs can only introspect from output tokens. You need hidden reasoning that is within the black box, self-knowing, intent, and motive to lie. I rather think accusing a
6. 246.▲
   by positron26 1y ago

   Reminds me of the "How did you take that picture?" Reddit meme. Who is the observer when we conceptualize of ourselves as deconstructed beings? Occasional self-awareness and reflection can be insightful, but when will you get ba