@Kantrowitz Really not impressed with Hinton’s reasoning. Consciousness is sensation, emotion, the essence of being that human intelligence sits underneath. If he wants to convince people computers feel those things he’s going to have to prove it.
Respectfully, I think @geoffreyhinton’s recent argument on multimodal models having subjective experience rests on a subtle but important shift in meaning.
He redefines “subjective experience” as a linguistic report about a perceptual malfunction (“what would be true if my system weren’t distorted”). The chatbot can output that sentence too, therefore it has the experience.
But this explains reporting behavior, not why there is something it’s like to have the experience (the hard problem). The words match, but the phenomenology may not.
Still admire the clarity and wit of the talk. Curious how others see it.
Apple, you love devs—we know it. NVIDIA GPUs in Macs now. Metal limits AI, CUDA rules. eGPUs aren’t real world. #AppleNVIDIA Even the hashtag looks good.
Here’s a product feature idea for ChatGPT: when you correct it and it acknowledges you are correct, it should incorporate this newly surfaced information in future answers. Not just repeat the same lies knowing now that they are lies.