I dunno, I guess I think it’s difficult to accurately guess really important details like the rate of increase. We agree that it’s fast, but why do lab leaders disagree by years on when AGI comes? There are huge implications of that because multiple years of singularity would result in very different worlds.
I think this article is mostly arguing against people who are too pessimistic by far so doesn’t really apply to the opposite point
There is proper thinking in this article, which I appreciate. However, the S-curve may still happen at a later date, potentially due to programming and/or government - or some limitation that hasn’t yet manifest itself.
In many cases, I’ve been frustrated by the inaccuracies that AI regurgitates regarding morality - and the errors it makes when it tries to help me build applications. It seems throttled, by design, dishonest.
Is there such a thing as an un-throttled AI? If not, it’s unfair to judge its performance today and it's impossible to predict its future.
My perspective is that both pure empiricists and pure extrapolators are naive. Exponential trends never continue indefinitely and so the real question is when and why does naive scaling end. "AGI" for example: Is that a matter of scaling up our current approaches, or are folks like Gary Marcus correct to think there is a qualitative element missing? No simple answer.
We tend to extrapolate farther than we should. A good example was aerospace, which went from the first jet aircraft to the first human in space in 20 years, to walking on the Moon just 8 years later. It would seem justified – conservative, even – to predict Mars in the 1980s and supersonic/orbital flight for everyday people. What was harder to see were the imminent technological and economic roadblocks that made naive scaling stop after Apollo. As a result the science fiction of the 70s and 80s is wildly over-optimistic.
I dunno, I guess I think it’s difficult to accurately guess really important details like the rate of increase. We agree that it’s fast, but why do lab leaders disagree by years on when AGI comes? There are huge implications of that because multiple years of singularity would result in very different worlds.
I think this article is mostly arguing against people who are too pessimistic by far so doesn’t really apply to the opposite point
The future has already happened & the past is yet to be, quant-wise.
There is proper thinking in this article, which I appreciate. However, the S-curve may still happen at a later date, potentially due to programming and/or government - or some limitation that hasn’t yet manifest itself.
In many cases, I’ve been frustrated by the inaccuracies that AI regurgitates regarding morality - and the errors it makes when it tries to help me build applications. It seems throttled, by design, dishonest.
Is there such a thing as an un-throttled AI? If not, it’s unfair to judge its performance today and it's impossible to predict its future.
My perspective is that both pure empiricists and pure extrapolators are naive. Exponential trends never continue indefinitely and so the real question is when and why does naive scaling end. "AGI" for example: Is that a matter of scaling up our current approaches, or are folks like Gary Marcus correct to think there is a qualitative element missing? No simple answer.
We tend to extrapolate farther than we should. A good example was aerospace, which went from the first jet aircraft to the first human in space in 20 years, to walking on the Moon just 8 years later. It would seem justified – conservative, even – to predict Mars in the 1980s and supersonic/orbital flight for everyday people. What was harder to see were the imminent technological and economic roadblocks that made naive scaling stop after Apollo. As a result the science fiction of the 70s and 80s is wildly over-optimistic.