4 ms·

Deceiving the board about ... Its investigation of misconduct? Sources and rights to training data? That the AGI escaped containment?

• 015a 3y ago
  Unhinged fringe take: They've already developed sparks of consciousness strong enough to create isolated, internal ethical concerns, but Sam suppressed those reports to push the product forward.
  • gizajob 3y ago
    The first word of your post being the most important part of it.
  • selfhoster11 3y ago
    Wouldn't be surprised if that was true. Public GPT-4 can be made to "think" using stream-of-consciousness techniques, to the extent that it made me rethink using insults as a prompting technique. I imagine that un-RLHF'ed internal versions of the model wouldn't automatically veer off into "as an AI language model" collapse of chains of thought, and therefore potentially could function as a simulator of an intelligent agent.