Yeah and the romans were 80% of the way to a steam engine. It only took another 2000 years to complete it.
OpenAI is "80% of the way" to AGI
21–30 of 35 posts
Re: OpenAI is "80% of the way" to AGI
#22AGI on static Neural Networks?! The list of requirements to have AGI is pretty long, and already that first one inconsistency makes the "hmmmm" bell sound.
Exactly. It is reasonable to believe that AGI will never be built on unchanging data, no matter how many times you iterate over it. Computers can and probably do already contain essentially all human knowledge, but without a continuous wave of new data to process and respond to then that surfer known as sentience will never have a wave to ride. Think of the comparisons to human minds. We are constantly assaulted with…
It is not a matter of sentience or consciousness - that is unrelated to this meaning of Intelligence (for AGI - "solving problems well"). It is that Intelligence implies continuous reflection, which brings continuous refinement of the set of ideas (of judgements) - which requires a dynamic DB.
Re: OpenAI is "80% of the way" to AGI
#23You changed the title and made it 93% more misleading. That completely made up number is meaningless. So's "AGI", for that matter.
> That completely made up number That «completely made up number» was said by "Chief research officer Mark Chen", and that is news - curious, and somehow interesting. It is in the article, and probably more notable than 'Inside OpenAI’s Reboot'. > Amid all this, company leaders believe they have arrived at the cusp of a milestone that could change the course of humanity: the creation of artificial general intelligenc…
It is almost certainly a completely made up number though. It's not quite on the same level of emptiness as "rate your emotion on a scale of 1-10", but it might as well be.
I.e. are we 80% of the way by the calendar, then if so, as measured by when? We're only about 66% of the way through the current year. Is it that they are 80% of the way through some final totally defined training or tuning procedure? Not likely. And it is 80% of the way to AGI, which is vaguely defined as that which can "outperform humans [by how much? what proportion of humans?] at most [how much, exactly, again?] economically valuable work [what work is not economically valuable?]."
Call a spade a spade, the 80% thing is 100% total bullshit (in the Frankfurtian sense, i.e. it is completely unconcerned about truth, and is all just about generating feels).
Re: OpenAI is "80% of the way" to AGI
#24Earlier quoted context omitted.
> That completely made up number That «completely made up number» was said by "Chief research officer Mark Chen", and that is news - curious, and somehow interesting. It is in the article, and probably more notable than 'Inside OpenAI’s Reboot'. > Amid all this, company leaders believe they have arrived at the cusp of a milestone that could change the course of humanity: the creation of artificial general intelligenc…
It is almost certainly a completely made up number though. It's not quite on the same level of emptiness as "rate your emotion on a scale of 1-10", but it might as well be. I.e. are we 80% of the way by the calendar, then if so, as measured by when? We're only about 66% of the way through the current year. Is it that they are 80% of the way through some final totally defined training or tuning procedure? Not likely.…
> Frankfurtian
It probably is, but not necessarily generally speaking: explicitly intended gross estimates may not be Galileian - they are, well, gross and estimates. You are pointing to the wrong detail: it is not that they would be "80% of the way": the news is that "the CRO estimates them at 80% of the way". It may be true that such is his estimation. And we cannot discard a priori his good faith. Of course more meaningful details would have had more juice.
But that he made that statement is actual banging news. It may just mean they are "80% on the way to redefine "AGI" in a biased partial exploitable way", but it is a statement that can make the journalists update the journals.
Re: OpenAI is "80% of the way" to AGI
#25From the author: "OpenAI’s charter defines AGI as “highly autonomous systems that outperform humans at most economically valuable work.” Its leaders won’t quite declare they’ve reached that threshold. But they no longer speak about it as a distant abstraction. Chief research officer Mark Chen estimated OpenAI is “80% of the way” to AGI. Brockman said that viewed from two years in the future, this may be remembered as…
Let us make it clear: "outperforming humans at most economically valuable work" is not AGI. AGI is when you implemented Intelligence (what intelligent people call "intelligence").
One example (of the many given many times) remains "adherence to truth": to get to the AGI they must have overcome the delioriousness issue - the outputs cannot be delirious. And another thing: if you can implement general Intelligence, you will also be able to achieve /optimized/ intelligence... As in: "here are all the theorems, in lean form, bringing to the output".
Re: OpenAI is "80% of the way" to AGI
#26Earlier quoted context omitted.
It is almost certainly a completely made up number though. It's not quite on the same level of emptiness as "rate your emotion on a scale of 1-10", but it might as well be. I.e. are we 80% of the way by the calendar, then if so, as measured by when? We're only about 66% of the way through the current year. Is it that they are 80% of the way through some final totally defined training or tuning procedure? Not likely.…
> Frankfurtian It probably is, but not necessarily generally speaking: explicitly intended gross estimates may not be Galileian - they are, well, gross and estimates. You are pointing to the wrong detail: it is not that they would be "80% of the way": the news is that "the CRO estimates them at 80% of the way". It may be true that such is his estimation. And we cannot discard a priori his good faith. Of course more m…
It could not be Frankfurtian if we say that he was just speaking colloquially and obviously only really meant something like "quite close", and indeed making up random percentages to stand for such vague feelings is really quite a normal thing in certain styles of everyday conversation. Whether or not such a style is appropriate in this context is debatable, but it could save him from the bullshit charge.
But I don't see how it being an estimation makes it less bullshitty, because there is simply no reasonable way to tie that 80% to anything concrete or meaningful. It is stated in such a way that evaluating the truth or accuracy of the estimate is basically impossible, i.e. it is bullshit by vagueness, even if it is an estimate. Only the "end of the year" claim is concrete, but the AGI definition has much of the same BS-via-vagueness problem, so it is still all very Frankfurtian to me.
So I don't see it as banging news, but just as more BS and almost certainly a nothingburger. But, hey, whatever, time will tell.
Re: OpenAI is "80% of the way" to AGI
#27Earlier quoted context omitted.
> Frankfurtian It probably is, but not necessarily generally speaking: explicitly intended gross estimates may not be Galileian - they are, well, gross and estimates. You are pointing to the wrong detail: it is not that they would be "80% of the way": the news is that "the CRO estimates them at 80% of the way". It may be true that such is his estimation. And we cannot discard a priori his good faith. Of course more m…
It could not be Frankfurtian if we say that he was just speaking colloquially and obviously only really meant something like "quite close", and indeed making up random percentages to stand for such vague feelings is really quite a normal thing in certain styles of everyday conversation. Whether or not such a style is appropriate in this context is debatable, but it could save him from the bullshit charge. But I don't…
> reasonable way to tie that 80% to anything concrete or meaningful
80% of the implementation of an internal project, I guessed.
Re: OpenAI is "80% of the way" to AGI
#28Earlier quoted context omitted.
Exactly. It is reasonable to believe that AGI will never be built on unchanging data, no matter how many times you iterate over it. Computers can and probably do already contain essentially all human knowledge, but without a continuous wave of new data to process and respond to then that surfer known as sentience will never have a wave to ride. Think of the comparisons to human minds. We are constantly assaulted with…
It is not a matter of sentience or consciousness - that is unrelated to this meaning of Intelligence (for AGI - "solving problems well"). It is that Intelligence implies continuous reflection, which brings continuous refinement of the set of ideas (of judgements) - which requires a dynamic DB.
Artificial general intelligence (AGI) is a hypothetical type of artificial intelligence that matches or surpasses human capabilities across virtually all cognitive tasks
I am saying that that without sentience, AGI will not happen, and that the current system does not allow for sentience any more than firecrackers make usable light.
Re: OpenAI is "80% of the way" to AGI
#29Earlier quoted context omitted.
It is not a matter of sentience or consciousness - that is unrelated to this meaning of Intelligence (for AGI - "solving problems well"). It is that Intelligence implies continuous reflection, which brings continuous refinement of the set of ideas (of judgements) - which requires a dynamic DB.
Artificial general intelligence (AGI) is a hypothetical type of artificial intelligence that matches or surpasses human capabilities across virtually all cognitive tasks I am saying that that without sentience, AGI will not happen, and that the current system does not allow for sentience any more than firecrackers make usable light.
If those cognitive tasks are the useful ones, those towards which we engineer things, then AGI will be a problem solver - you do not need sentience for a problem solver (like you do not need it for a calculator).
Re: OpenAI is "80% of the way" to AGI
#30Claiming AGI when they don't even have the best performing LLM out is a bit bold :p
OpenAI doesn't have the best-performing public model, but it's an open secret that Astra, the internal model, is better than anything Anthropic has internally.
For at least another week!!1!
The frontier race is probably very tight actually..