zlacker

[parent] [thread] 3 comments
1. 0xbadc+(OP)[view] [source] 2023-05-23 13:05:18
We made the world's fastest plane and got to the moon in a span of about 10 years. Since then, nothing remotely as groundbreaking has occurred in flight or space travel. That was 54 years ago.

180 years ago the electric car was born, and about 105 years ago electric cars were the most popular car. Electric cars are finally becoming popular again but are still dwarfed by internal combustion.

Progress isn't linear, and the last 10% takes 90% of the effort. Without a very specific and heightened pace of work, the work will become monotonous and innovation will drag to a halt. It's not that we can't make groundbreaking work, it's that it's much harder and more expensive than we care for.

replies(1): >>cubefo+u7
2. cubefo+u7[view] [source] 2023-05-23 13:41:02
>>0xbadc+(OP)
If anything, progress speed in AI has greatly increased. The AlexNet architecture combined CNNs and backpropagation, and used GPUs, none of which was new at the time. In contrast, GPTs use an entirely new architecture, the transformer. Progress is accelerating, not slowing down.

By the way, space was stagnant because there was little commercial investment for decades. Most development was done by NASA themselves (space shuttle) which turned out to be highly ineffective. But now there is a lot commercial activity. And relatively groundbreaking progress has occurred already: SpaceX has built a reusable lower stage for their medium lift rocket, which strongly reduces their launch cost. As a result of this cost reduction, they are currently building a global satellite internet network in low orbit with low latency and high speed. And they are also working on a rocket which has a similar payload capacity to the Saturn V, while being fully reusable. The cost will be a tiny fraction.

In the field of AI, too, the commercial research is currently taking off. Microsoft is pouring billions into OpenAI, and Google is racing to keep up.

Nothing points at a slow down.

replies(1): >>0xbadc+wZ
◧◩
3. 0xbadc+wZ[view] [source] [discussion] 2023-05-23 17:16:04
>>cubefo+u7
Progress cannot accelerate infinitely. Literally nothing in the universe does. It accelerates until it slows down again. It's going to slow down again, and we won't reach anything like what these claims pretend in 10 years, or even 20.

Most technological progress is based on one of two things: 1) obvious commercial viability combined with a strong economy, copious capital, and a novel business solution that saves labor cost+time, or 2) warfare. We're not in a large-scale war right now, so that one's out. The economy is slowing down, and we're on the down-swing of the hype cycle for AI, seeing as everyone has bought something called "AI" but there's no new business value. Without a real business case or a decent war, the investment in innovation is going to falter.

The 3rd way we get technological innovation is after decades of very slow incremental progress. That's what led to GPT. But just because decades of research eventually lead to something useful doesn't mean it's immediately going to lead to yet more utility in the near future.

Another example: the elevator. The parts that comprise a simple elevator - a rope, pulley, and ratchet - have all existed for over 2,000 years. And for 2,000 years, people have wanted to lift heavy things high in the sky. But it was always dangerous because the rope would unexpectedly break, so nobody used them. Until one day, some random guy in the 19th century combined both a ratchet and a rope+pully, and suddenly it was safe to lift things very high. All it took was one moment for something we could already have done before to become a viable product.

That was 171 years ago. So how come that one innovation didn't shortly lead to a space elevator? We've known since 1895 that it's possible, and as far as we can tell it seems pretty straight forward how to do it. But still nobody's done it. Why? Not because it's impossible, but because there simply isn't enough money and will to do it. Progress doesn't happen just because it can happen.

Doesn't matter if you believe me or not. Check back at this thread in 10 years and tell me I'm wrong then. I won't be.

replies(1): >>cubefo+kc1
◧◩◪
4. cubefo+kc1[view] [source] [discussion] 2023-05-23 18:02:54
>>0xbadc+wZ
Well. The forecasting community estimates general intelligence (incl. robotics and passing a 2 hour Turing test) for 2031.

https://www.metaculus.com/questions/5121/date-of-artificial-...

I don't know what would change your mind.

[go to top]