AI is not “smarter” than human and won’t be for a few more years minimum.
I wish I still had the link to the document that showed the various defense agencies and their research priorities. Probably more money spent on AGI R&D than the rest of the world combined. They’re even doing some Neuralink-type stuff.
I don’t know if that is preferable to shutting AGI down outright.
You personally can get access to GPT-4 which exceeds human ability on most tests. For the sake of argument, say that this still doesn't mean it's smarter than humans. Its training was finished nearly a year ago, and was nerfed before it was released to the public. Even if you think the pre-nerfed GPT-4 was still not smarter than humans, do you think the next version won't be? And you think the next version hasn't already been trained? If you think one smarter than humans doesn't exist yet, do you think that you will be immediately informed when it's created?
GPT4 is incapable of sharing opinions. If that is a technical limitation then that should answer your question. Do any of the other LLM models hold consistent opinions? I’m going to guess that’s a no.
(Edit: also see https://twitter.com/MaartenSap/status/1587554332316119040)
GPT4 also cannot store and retrieve memory sequentially and iteratively in real time like humans can. There is a hard limit of a few pages of text. Then poof, memento. Sure, those memories can be synthesized over the next few months in a very costly process that starts with reading your chat histories. But it needs another breakthrough-level advancement before that’s on par with humans. So we are still some years off.
> AI is not “smarter” than human and won’t be for a few more years minimum.
AI has collective intelligence served up individually. This is something that humans as individuals have the inverse of. Humans have to come together to arrive at hive mind, while AI is trained on hive mind, and is as a result "smarter" than individual humans in terms of baseline intelligence.
Additionally the speed with which AI can learn new things supersedes that of humans (e.g., learning new languages, reading vast amounts of information and summarizing it, remembering that newly learned information in its totality when reasoning about novel tasks).
This is to say, AI is already smarter than the average human, today. In a few more years we are facing AGI. Further, we have no idea how to tell when the threshold's been crossed. We can postulate. However AI is learning all of those guesses via the training data.
Hence many smart people are suggesting a pause, for good and legitimate reason. We as a society are not ready.
No. Shut it down. People are by an large ignorant. They will ruin society voting for things unearned and over power rational minds. Shut it down and create a voting system where the needy have almost no input. The needy will destroy any civilization if given the means. Kill it. I know I will get many down votes but that is fine. Either kill it or it will destroy everything. Want everything destroyed then go right ahead.
You won’t get downvoted because this thread got buried, despite the article being written by the literal pioneer of AGI (full disclosure, also a longtime friend)
If AI doesn’t kill us, something else likely will end things in our lifetimes. But AI is the only thing that can avert those other disasters, unless we can cure the idiocracy we got ourselves into. To think that Neuralink is the best thing Elon could come up with to that end.
This was written by one who has staked his reputation on making AGI. I doubt he will ever admit that someone makes it before he does it. He will just say the other one is not AGI but the one he is working on will be 'true' AGI.
9 comments
[ 3.5 ms ] story [ 14.4 ms ] threadI wish I still had the link to the document that showed the various defense agencies and their research priorities. Probably more money spent on AGI R&D than the rest of the world combined. They’re even doing some Neuralink-type stuff.
I don’t know if that is preferable to shutting AGI down outright.
You personally can get access to GPT-4 which exceeds human ability on most tests. For the sake of argument, say that this still doesn't mean it's smarter than humans. Its training was finished nearly a year ago, and was nerfed before it was released to the public. Even if you think the pre-nerfed GPT-4 was still not smarter than humans, do you think the next version won't be? And you think the next version hasn't already been trained? If you think one smarter than humans doesn't exist yet, do you think that you will be immediately informed when it's created?
GPT4 also cannot store and retrieve memory sequentially and iteratively in real time like humans can. There is a hard limit of a few pages of text. Then poof, memento. Sure, those memories can be synthesized over the next few months in a very costly process that starts with reading your chat histories. But it needs another breakthrough-level advancement before that’s on par with humans. So we are still some years off.
https://twitter.com/bengoertzel/status/1639378920465395712
I’m sure if the maestro himself were to chime in here he would be able to write a full book (literally) on why GPT4 is not AGI.
AI has collective intelligence served up individually. This is something that humans as individuals have the inverse of. Humans have to come together to arrive at hive mind, while AI is trained on hive mind, and is as a result "smarter" than individual humans in terms of baseline intelligence.
Additionally the speed with which AI can learn new things supersedes that of humans (e.g., learning new languages, reading vast amounts of information and summarizing it, remembering that newly learned information in its totality when reasoning about novel tasks).
This is to say, AI is already smarter than the average human, today. In a few more years we are facing AGI. Further, we have no idea how to tell when the threshold's been crossed. We can postulate. However AI is learning all of those guesses via the training data.
Hence many smart people are suggesting a pause, for good and legitimate reason. We as a society are not ready.
If AI doesn’t kill us, something else likely will end things in our lifetimes. But AI is the only thing that can avert those other disasters, unless we can cure the idiocracy we got ourselves into. To think that Neuralink is the best thing Elon could come up with to that end.
Not a fun situation to be in.