Hacker News Reader

Comment by untog | Hacker News Reader

> did the company and the investors ever actually believe that the robotaxi stuff was even close to viable

IMO, absolutely. The hype for self driving cars a few years ago was massive. I forget who, but a prominent SV investor offered to make a bet with someone that self-driving taxis would be on the road in five years. There was plenty of it in HN, too.

That's the investors. Did Travis think so too? The insane amounts of money they spent on acquisitions and developing the tech suggests yes: they were trying to be first to a huge potential market. But maybe he was doing all that to scam investors, who can tell.

View on HN

I think all the self-driving cars hysteria was just momentum from the hype of deep learning advances. I remember thinking about the idea of self driving cars within 10 years, "this seems crazy, but I guess I've been wrong before...". A few years later it seems even crazier. But I've also been wronger than that...

There were (are?) people talking about Artificial General Intelligence being just around the corner, which is a sign of people buying way too much into the hype cycle.

It's the sort of problem that is easily prone to extreme optimism extreme pessimism cycles. You get 90% of the way there and then the difficulty skyrockets. Getting to 95% is as hard as going from 0 to 90%. Getting to where you need to be for it to be mainstream is like 10-20X as hard as making an impressive looking proof of concept.

Space flight is another thing like that. We were supposed to have manned flights to the outer solar system by 2000. We're getting reusable first stages that are actually economical to fly and commercial orbital spacecraft in 2020. That was supposed to be 1980s stuff according to 1950s and 1960s space hype.

Putting someone on the moon? 1969. But that was done with a monstrously expensive finicky one-off architecture that could never be anywhere close to economical, and it was dangerous as hell. Routine trips to the moon that could even approach economic sustainability are at least 10X as hard as Apollo.

I strongly suspect safe, sustainable, and affordable nuclear fission power is yet another one. Getting some atoms to smash and boil some water? Easy! Build some power plants? A little harder. Then you hit the edge of that bath tub curve and the problems (waste, safety, fuel cycle, cost management) multiply and it gets brutally hard really fast.

We should learn to recognize engineering problems that have this kind of "bath tub curve" for difficulty vs ones with gentler learning curves.

I like these analogies to other engineering challenges!

It's easy to make predictions when outside a field (and I'm guilty of this fallacy too), but easy to miss the true challenges. For example, a lot of people just assume that fusion reactors will be better than fission, because that's what they were told their entire life and because we still haven't figured it out. But looking at energy density or thermal transference and fusion just looks like a really crappy heat source, something that would be incredibly difficult to ever engineer into a product without a secondary invention that is just as difficult, if not more, than the initial idea of fusion.

Fusion is much better on safety, waste, and fuel cycle. The thing I wonder is whether it can be made cost competitive with renewables plus storage when the latter is experiencing falling costs at scale. At midday solar is already cheaper than most other power sources, and economical storage at scale is probably a much easier problem than fusion (or really good scalable fission for that matter).

I work in machine learning, and I've noticed a particular dynamic in how people treat predictions like this.

In general, when there is a breakthrough in a field of research—like what deep learning went through recently—there is a gold rush of new projects that push the boundaries.

A certain category of people look at these new startups, and project that momentum out to some extremely distant problem (like AGI or fully autonomous vehicles), without understanding if this field of research necessarily leads to solutions to those problems, at least in linear fashion.

Then there is another category of person who looks at those predictions and dismisses the entire field as hype, ignoring the surge of valuable projects that kicked it all off.

Totally agree, and as newcomers enter the field they can have a mistaken sense of time as they digest a bunch of really amazing new ideas, like has happened with deep learning in the past decade (the decades before that felt much slower to me).

Reply on news.ycombinator.com