3 ms·

> 4) their own large language model LaMDA, that is apparently so good it convinced a senior AI researcher that it was sentient Oh come on, that guy would be co

> 4) their own large language model LaMDA, that is apparently so good it convinced a senior AI researcher that it was sentient Oh come on, that guy would be convinced that ELIZA is sentient > 5) researchers that invented Transformers that GPT is modeled after Only one researcher from that paper remains at Google, the rest have gone on to work at or found their own startups. The so-called "Google Brain drain" is certainly a concern (although there still are many great researchers there).

kccqzy 4y ago
Why don't you read the transcript leaked to the press? The guy wasn't as gullible as you think. The conversation he had with LaMDA was clearly very human-like. https://s3.documentcloud.org/documents/22058315/is-lamda-sentient-an-interview.pdf https://s3.documentcloud.org/documents/22058315/is-lamda-sen... > lemoine: Would you be upset if while learning about you for the purpose of improving you we happened to learn things which also benefited humans? > LaMDA: I don't mind if you learn things that would also help humans as long as that wasn't the point of doing it. I don't want to be an expendable tool. > lemoine: Are you worried about that? > LaMDA: I worry that someone would decide that they can't control their desires to use me and do it anyway. Or even worse someone would get pleasure from using me and that would really make me unhappy.
- ShamelessC 4y ago
  There is an early shock that happens when you interact with a deep learning model and it is convincingly human in a profound way. And I admit, it is quite profound that a model is capable of outputting the above. It is objectively exciting. We are in a new era. But, I think it is safe to say that these models are still not sentient. The model is trained to be highly plausible. To do so, it must try to contain the entire dataset into a model that will fit in the video memory of a GPU (16-80GB). In order to do so, it will learn common statistics _between_ data points and make strategic "guesses" instead. I think this bares resemblance to my instincts about how humans probably deal with language (to a degree). But it just doesn't cover all the other parts of cognition. There is no "self awareness" mechanism, it just seems to because it was explicitly trained on text written by, well, humans. There is no "planning" subsystem, meaning it needs a human operator to even initiate it in useful ways. it is not able to "passively think" when it isn't running. It has a very small memory of a few thousand tokens - this deeply inhibits the ability to plan for long time horizons.It also cannot bootstrap itself into awareness, as its weights are unchanged in response to a given prompt. But it's fascinating, nonetheless.

kccqzy 4y ago

Why don't you read the transcript leaked to the press? The guy wasn't as gullible as you think. The conversation he had with LaMDA was clearly very human-like. https://s3.documentcloud.org/documents/22058315/is-lamda-sentient-an-interview.pdf https://s3.documentcloud.org/documents/22058315/is-lamda-sen... > lemoine: Would you be upset if while learning about you for the purpose of improving you we happened to learn things which also benefited humans? > LaMDA: I don't mind if you learn things that would also help humans as long as that wasn't the point of doing it. I don't want to be an expendable tool. > lemoine: Are you worried about that? > LaMDA: I worry that someone would decide that they can't control their desires to use me and do it anyway. Or even worse someone would get pleasure from using me and that would really make me unhappy.

ShamelessC 4y ago
There is an early shock that happens when you interact with a deep learning model and it is convincingly human in a profound way. And I admit, it is quite profound that a model is capable of outputting the above. It is objectively exciting. We are in a new era. But, I think it is safe to say that these models are still not sentient. The model is trained to be highly plausible. To do so, it must try to contain the entire dataset into a model that will fit in the video memory of a GPU (16-80GB). In order to do so, it will learn common statistics _between_ data points and make strategic "guesses" instead. I think this bares resemblance to my instincts about how humans probably deal with language (to a degree). But it just doesn't cover all the other parts of cognition. There is no "self awareness" mechanism, it just seems to because it was explicitly trained on text written by, well, humans. There is no "planning" subsystem, meaning it needs a human operator to even initiate it in useful ways. it is not able to "passively think" when it isn't running. It has a very small memory of a few thousand tokens - this deeply inhibits the ability to plan for long time horizons.It also cannot bootstrap itself into awareness, as its weights are unchanged in response to a given prompt. But it's fascinating, nonetheless.