Comment by ilovefood | Hacker News Reader

Comment by ilovefood | Hacker News Reader

Discrediting work being done by others in the realm of self driving cars is rarely a good segway for introducing a company that automates checkout-free stores powered by machine learning / image recognition (which the articles builds up to). Sure, the developments have been quite disappointing for the cars but that's a way bigger challenge that currently has examples running. And what is the source of that disappointment? "X said it should be there in 3 years and it's not so it's a fail", really? That strikes me as dismissive and at best counter productive. Marketing and developping are two different things, don't use the shortcomings of the former to discredit the latter. It doesn't help anyone, especially new people joining the field.

> "The deep learning hype is now deflating pretty strongly too, with even the biggest "enthusiasts" now admitting that things ain't as rosy as they were hoping."

This comes out as bad faith, so a reference would be needed... Andrew NG was pointing at the use of wrong/bad data, and not the models or approach per se...

So yeah, crapping on a trendy topic (deep learning) to market your venture that ironically itself uses that trendy topic (i quote "artificial intelligence"), is at best totally uncool...

I read it in quite the opposite direction, the cautionary note here is that if you overpromise and under deliver you might be able to secure early funding but eventually you precipitate yet another AI winter.

The author, as someone in the space, is cautioning for more realism which arguably is lacking in public communications within that field.

I think his last paragraph makes that very clear and echoes a conversation I was having with someone last week. There are a lot of practical if often boring wins to be had. We're just probably not going to get some of the things like "never need to own/drive a car" that had some people so excited.

As the piece articulates, there has been a big deflation of many high expectations from "just around the corner" to "maybe in 40 or 50 years."

Has there? I can't imagine investors/corporations dumping billions, annually, into an project which won't produce a return for 40-50 years.

>Has there?

Yes? That's what the whole article is about.

To be clear, it's doubtless brought benefit to assistive driving systems. And we may even see full autonomy under limited conditions (e.g. specific limited access highways in good weather) in a significantly shorter period of time.

But door to door robo taxis? Seems highly unlikely.

>I can't imagine investors/corporations dumping billions

There have been big pullbacks. And I wouldn't be the slightest bit shocked if Google/Alphabet pull the plug on Waymo one of these days--especially if they have to start tightening their belt for some reason.

> Yes? That's what the whole article is about.

Articles like this have been written for 5 years now, but the money keeps flowing into these projects. Argo, Aurora and Aptiv etc. are all still burning mountains of cash.

I am shocked the industry didn't take Uber/Lyft selling off their self-driving divisions as the signal to stop funding these efforts.

Lots of stupid money flowing around these days.

I'm genuinely curious how the various self-driving efforts broke down among:

- We can do it!

- Maybe we can't do it but even partial success makes it worthwhile

- What a pipe dream but it would be morally wrong not to separate greedy VCs from their cash

I do think it's different from 5 years ago. Yes, some researchers were throwing cold water on the idea then but they were seen as contrarians. And, on boards like this one, you'd have no shortage of people going but Waymo is going to have a taxi service next year! Today, it's closer to being accepted wisdom.

Ha! I kinda stopped paying attention to most consumer-oriented startups a few years back when some kooky poetry delivery service got a few million in funding. (I am trying to find a link to it). Then there was that "pizza robot" company ...

I get that VCs will fund stupid things like that to grow entrepreneurs and build a portfolio of companies that maybe grow enough to get acquired and produce some profit. But funds/corps dumping $250 million+ annually into self-driving seems crazy. The opportunity costs of that alone are outrageous especially when there are ample and ripe opportunities to disrupt "traditional" businesses and business models across all kinds of market verticals.

> I do think it's different from 5 years ago. Yes, some researchers were throwing cold water on the idea then but they were seen as contrarians. And, on boards like this one, you'd have no shortage of people going but Waymo is going to have a taxi service next year! Today, it's closer to being accepted wisdom.

Yes, and this has always been perplexing. I work at an AI startup which actually uses it to accomplish rather straightforward tasks - and it's really hard to perfect with the minimal tolerances for error that we have to adhere to. When you start to scale the problem up to self driving it becomes evident quickly that it will take a looong time to get the level 5; yet people in this business were still insisting it was around the corner.

I'm by no means an expert but a number of years ago I saw a presentation by MIT's John Leonard (he was involved in one of the early DARPA contests). One of the interesting things he showed was dash cam footage he had taken over the course of about a week commuting from Brookline where he lived to MIT. And he pointed out all the things that would be really difficult to do.

As a non-expert but a very longtime driver this made an awful lot of sense to me. But so much money and brains were saying success was right around the corner that I was half-convinced that I and other skeptics were missing something.

I think, in addition to the scammers and dumb money, it's just that a lot of people who should have known better just looked at the pace of advance over the previous 5+ years and figured "How could we not iron out the remaining kinks in a few more?" Add to that the number of people who don't like cars much and just so desperately wanted this future where they never needed to own a car or drive again.

Look at the actual numbers though. I think the biggest writing on the wall that the hype is starting to be weighed in valuation is Waymo.

Waymo is arguably the most technologically advanced of all the self-driving efforts. However they were initially valued at 175 billion, Morgan Stanley downgraded it to 105 billion, and their last external funding round valued them at 30 billion.

You don't need to be a VC fund to realize that is a hell of a drop (most probably because the initial valuations were hyped by the 'just around the corner' momentum).

> Andrew NG was pointing at the use of wrong/bad data, and not the models or approach per se...

Except that Deep Learning is equal parts data & model. And he wasn't saying that the data was bad, he was saying that it turns out that the data collected turned out to be so context-dependent that it wasn't usable with even a small change in that context. That's not "bad data", that's "the real world".

Reply on news.ycombinator.com