New Game: Take a shot every time I say dot product. Additionally, the AGI part is pure speculation and opinion. To clarify what I meant in the video, GPT4, I don't think, is sentient but going from 100+billion parameters from GPT3 in a couple years to 10x to 1 trillion, we can predict GPT5 will be around 10 trillion, and GPT6 will be 100 trillion which would be comparable to a human brain
One correction, it is a rumour that GPT-4 has 100 trillion parameters. It was never revealed and it is very unlikely that a 100T model can be put to practical use.
Great video! I’ve been meaning to watch Karpathy’s tutorial but haven’t had the time to sit down with to watch the full thing yet. You covered most of the technical material I was missing in a much shorter video so I appreciate it. I have some qualms with your claim that some version of GPT may be conscious/sentient, although that may just come down to definitions of terms, so I’ll let you have it :P Thanks, subscribed!
That's fascinating. I have to admit to being skeptical that an LLM by itself could lead to AGI, but I am blown away by the test taking and coding it can do. For sure I can see it as a module in an AGI that also has agency and imagination.
Thanks for the video! It's nice to learn more about how GPT works more in-depth, as the main knowledge I have with NN's is 3blue1brown's series on it. Thought I think I didn't understand your last point about consciousness: At first it seemed you argued that because it can describe consciousness it must be conscious, but then I thought you actually meant that whatever GPT's description of consciousness is a definition of its consciousness? Like better, more insightful explanations imply more consciousness of some sort?
No sure why you confuse inelegance with cautiousness while cautiousness barely has any solid definition, if intelligence is a spectrum why we don’t see that in cautiousness too?
New Game: Take a shot every time I say dot product. Additionally, the AGI part is pure speculation and opinion. To clarify what I meant in the video, GPT4, I don't think, is sentient but going from 100+billion parameters from GPT3 in a couple years to 10x to 1 trillion, we can predict GPT5 will be around 10 trillion, and GPT6 will be 100 trillion which would be comparable to a human brain
One correction, it is a rumour that GPT-4 has 100 trillion parameters. It was never revealed and it is very unlikely that a 100T model can be put to practical use.
Great video! I’ve been meaning to watch Karpathy’s tutorial but haven’t had the time to sit down with to watch the full thing yet. You covered most of the technical material I was missing in a much shorter video so I appreciate it. I have some qualms with your claim that some version of GPT may be conscious/sentient, although that may just come down to definitions of terms, so I’ll let you have it :P Thanks, subscribed!
THANKS. I love your channel.
That's fascinating. I have to admit to being skeptical that an LLM by itself could lead to AGI, but I am blown away by the test taking and coding it can do. For sure I can see it as a module in an AGI that also has agency and imagination.
Thanks for the video! It's nice to learn more about how GPT works more in-depth, as the main knowledge I have with NN's is 3blue1brown's series on it. Thought I think I didn't understand your last point about consciousness: At first it seemed you argued that because it can describe consciousness it must be conscious, but then I thought you actually meant that whatever GPT's description of consciousness is a definition of its consciousness? Like better, more insightful explanations imply more consciousness of some sort?
I'm not a good comedian but I didn't need to be called out like that today, no. 😂
Over a hundred TRILLION parameters?🤔
No sure why you confuse inelegance with cautiousness while cautiousness barely has any solid definition, if intelligence is a spectrum why we don’t see that in cautiousness too?
consciousness is 'con-science' or 'agreeable sensations'
intelligence is 'inter-lego' or 'read between'
the two are distinct
Hey bro your doing good job. Heads up there is software that makes your eyes straight even if your lookimg at your notes. Forgot what is call . Peave