Hacker News Reader

Comment by roymurdock | Hacker News Reader

Good behind the scenes analysis of why Waymo's case was flimsy here from Sarah Jeong, a lawyer and journalist who has been live tweeting the trial:

https://www.theverge.com/2018/2/8/16993208/waymo-v-uber-tria...

Summary - Waymo was claiming Uber violated 8 trade secrets (down from 100s), and had only directly discussed these trade secrets for 45 minutes over the past week. "Can you really explain eight self-driving car trade secrets in 45 minutes?"

"And then there’s the part where Anthony Levandowski is not on trial, Uber is. Even if Levandowski took 14,000 confidential documents, that doesn’t mean that Uber did something wrong — there isn’t a clear link to how those documents got onto Uber computers or were used in Uber self-driving cars."

View on HN

The lawyers did such a bad job. They bungled the litigation. Now it seems like they have endorsed a strategic option (settlement) that was way too risk averse.

Alphabet had already won the PR war. They had alot to gain by grinding Uber into the dirt. Now Alphabet comes off looking like a paper tiger for a settlement that requires Uber to do very well to have any value. Let's not forget they are a major shareholder in Lyft. I can imagine that Larry Page must not be very happy right now.

> Now Alphabet comes off looking like a paper tiger for a settlement that requires Uber to do very well to have any value.

The value of the settlement for Alphabet is not in the stock, it is in the “agreement to ensure that any Waymo confidential information is not being incorporated in Uber Advanced Technologies Group hardware and software.”

Of course, the details of that aren't going to be shared, but that's the thing matters.

> Let's not forget they are a major shareholder in Lyft.

Are they? I know GV has a significant stake in Uber, and Waymo has a cooperative deal with Lyft, but I don't recall Alphabet or any of it's subsidiaries holding a big share of Lyft.

Edit: as pointed out in the response, Alphabet is invested in Lyft through CapitalG.

https://www.theverge.com/2017/10/19/16503742/lyft-capitalg-g...

I'm not sure how much I would trust any promise from Uber to not continue to use technology they already have, even if it was stolen. Uber doesn't seem the company to just say "Ok" and delete the information that wasn't theirs. I would be skeptical a lot of under the table development is still in action, but with a better mindset of not getting caught in place.

I'd imagine Uber's consent to not use waymo trade secrets comes with Uber agreeing that the points in contention are waymo trade secrets. Doesn't that mean waymo has a very good standing if Uber were to use these trade secrets?

I mean this is better than a patent because a trade secret will never expire in perpetuity, right? I also imagine this extends to anyone acting on Uber's behalf?

I'd imagine that the commitment to Uber and Waymo working together to assure that Waymo trade secrets are not used that is referenced in the settlement announcement involves more than Uber promising and Waymo accepting assurances at a distance.

According to the Techmeme headline Uber pledges not to use Waymo trade secrets: "Uber settles Waymo lawsuit, pledges not to use Waymo's trade secrets; Waymo will receive 0.34% stake in Uber worth about $245M"

So how is this not a win for Waymo?

Uber has a significant documented history of flouting both the law and agreements they've made?

1. DK has clearly been telegraphing that they are no longer accepting practices of flouting agreements, no?

2. Isn't that exactly what this settlement is intending to do? It sets the legal precedence for "don't skirt this one, we're serious about it".

The real win for Waymo is what happened to Levandowski and Kalanick after they filed the complaint.

Uber lost two key employees and paid Waymo for the trouble.

Waymo/Google counsel earned their annual bonus.

Is it really? I would think Waymo's primary goal is making money for their shareholders, not pursuing personal vendettas.

I would imagine that weakening a competitor in their space would be decently correlated with expected return for shareholders. Not sure where you got the idea that the only benefit would be "personal".

Sure, but I have a hard time imagining that "what happened to Levandowski and Kalanick", as individuals, is going to weaken Uber in the marketplace. If anything, going from Kalanick to Khosrowshahi might have actually helped Uber in the long run.

Yeah, I agree with this. Both Levandowski and Kalanick seem to be incredibly toxic people; removing or lessening their influence on Uber, and replacing them with decent people, would only seem to strengthen Uber's position and reputation going forward.

Which interestingly circles back to another comment that was made:

> a settlement that requires Uber to do very well to have any value.

Keeping employees from straying too close to the line of 'sharing trade secrets with their next employer' makes money for shareholders. There's a chilling effect.

The next Levandowski will think twice before following in this one's footsteps - even if their actions would be 100% legal. The next Kalanick will think twice before hiring him.

Levandowski can't be all that smart if he thought he could download 14,000 documents onto his work laptop, without Google noticing. I mean, hello? Any sane I/T security protection architecture would assume that despite the best security you can put on employee's laptops and mobile devices, a sufficiently determined criminal or state actor is going to be able to get onto the laptop.

So having extensive monitoring on work machines is not because you don't trust your employees --- but because you don't trust that the employee's laptop / mobile device is always going to be under the control of the employee. And if it does get compromised, you want to be able to state with confidence how much data might have gotten compromised, or how much access the criminal or state sponsored attacker might have been able to gain access to.

The fact that it can also be used to find incriminating evidence for back-stabbing employees who are intent on screwing over their employer is just a bonus.

And if he wasn't smart enouh to realize this, I wonder how much Waymo lost and how much Uber gained when they lured him away from Waymo. :-)

AFAIK the theft was only noticed after supplier noticed that Otto was ordering similar parts as Waymo did, hence notified them.

It's meaningless if Uber didn't in fact possess any Waymo trade secrets. You're not allowed to use other people's trade secrets anyway.

Which implies that Uber needs to start off it's plans to build driverless cars from zero or partner, which is precisely what the lawsuit was about. It's also what Otto's acquisition price was predicated on - that it had built IP from zero for driverless cars. The settlement is an indication that it did not do that, which means they'll likely have to scrap a large piece of it's incremental work.

> You're not allowed to use other people's trade secrets anyway.

Being allowed to do something and actually doing something are two very different things.

What? Nothing is implied. No one is starting over. Money changes hands and everyone saves face.

> pledges not to use Waymo's trade secrets

If this statement is indeed true and Uber isn't totally brazen, then yes it would imply that Uber is going to have to seriously clamp down on how it uses the data/IP they got from Otto (which clearly they got from Google/Waymo). Not a lawyer, but that's how I read it.

They didn't get any IP from waymo, that is why waymo settled. There is no additional clamping down, they just need to continue to show alphabet that they're not doing anything fishy.

See my point above. You don't settle if your claim is 100% baseless, so it's very unlikely that no trade secrets were obtained.

> Which implies that Uber needs to start off it's plans to build driverless cars from zero

What does this mean in a practical sense? rm -rf /car? Can the Uber engineers who worked on this continue to do so, or are they "tainted"?

> The settlement is an indication that it did not do that,

Not really. The settlement is an indication that someone did an ROI calculation and decided that .043% of Uber is cheaper than the cost of seeing the trial to conclusion with the possibility of losing.

Based on what we all read in the news today, winning the autonomous race against a competitor is a double digit billion dollar opportunity. IMHO, the 0.043% is nothing more than a signal, rather than an absolute ROI. That number is a rounding error on a Alphabet balance sheet. If this was Oracle/SAP/Microsoft it would be a different story, where punitive damages are part of it's licensing model due to lack of growth from existing products.

> The settlement is an indication that it did not do that, which means they'll likely have to scrap a large piece of it's incremental work.

That isn't true. Uber didn't use any trade secrets from waymo, thats why waymo settled. The point of the oversite is to continue to show that uber isn't using their trade secrets. There is no part of the uber tech stack that needs to be retired as an outcome of this case.

> Uber didn't use any trade secrets from waymo, thats why waymo settled.

Waymo has plenty of motivation to settle if they could get adequate behavioral commitments in the settlement, even if they thought Uber had and had used Waymo trade secrets. We don't know here details of the behavioral commitments around preventing use of Waymo technology, but their existence was in the public release.

Given that, if Waymo wins the self-driving race, Uber is a likely customer and is a company Alphabet already owns a nice chunk of through GV, there is plenty of incentive to minimize cost to Uber as well as Waymo while mitigating any risk of Uber using Waymo tech. Dragging out the court process isn't in anyone's interest, whether or not Uber stole and used Waymo technology.

> To be clear, while we do not believe that any trade secrets made their way from Waymo to Uber, nor do we believe that Uber has used any of Waymo’s proprietary information in its self-driving technology, we are taking steps with Waymo to ensure our Lidar and software represents just our good work. [0]

> Uber didn't use any trade secrets from waymo, thats why waymo settled.

This is still unclear.

> There is no part of the uber tech stack that needs to be retired as an outcome of this case.

Honestly, no one knows for sure, everything here (including my statements) is pure conjecture. Here's the reality though - Uber did not explicitly say "we absolutely 100% did not steal trade secrets". Outside the court of law, that tells me they are going to lose something of value...to what extent, I guess only Uber knows.

[0] https://www.uber.com/newsroom/uber-waymo-settlement/

You're not allowed to use other people's trade secrets anyway

Sure you are, you're just not allowed to reveal them. That is, once they're revealed [to outsiders], they can be used by anybody. The only cause of action that I know of is to go after the revealer (as we have here), nobody who learns (and/or uses) anything from a disclosure of trade secrets has anything to worry about.

That's not really true. If learn a trade secret you cannot use it. If it became public knowledge through no fault of your own, it's no longer a trade secret.

Put if you steal the recipe to Coke and email it to Pepsi, they can't use it, even if use doesn't disclose it.

>If learn a trade secret you cannot use it.

It depends on how you learn the a trade secret.

The law in California is complicated, and appears to hold that a company is liable if they "knew or should have known" the information they were using was misappropriated. This came up a bunch back in the SCOX v. IBM days.

Does that mean Uber is valued at $72 billion?

The article is not a very comprehensive look at what's going on. The thread running through the comments - that the case Waymo had against Uber was flimsy - is questionable.

You don't toss someone a few hundred million dollars because your case is 'flimsy', and 'winning or losing' in court is too simplistic an analysis of the impact of the lawsuit on both sides as well as their third party competitors.

The Jeong article reads like someone read Uber's submissions to the court, phrased them in common vernacular, then ignored everything that didn't fit their narrative.

You do toss someone a couple hundred million in private equity for an immediate end to the damage to the company's brand, legal fees and recruiting efforts.

If Uber was in any real jeopardy of being found liable here they would have settled for far more. We know this because if Uber lost the case outright they would face an injunction against their self-driving business that would probably force them to license Waymo tech or put an end to the whole endeavor. Uber ATG is worth a lot more than what they paid in equity, and this is not an amount that matters at all to Google. The dollar figure is just there so Google can say they didn't come away with nothing.

> Uber ATG is worth a lot more than what they paid in equity, and this is not an amount that matters at all to Google. The dollar figure is just there so Google can say they didn't come away with nothing.

The settlement wasn't just for money, it also barred Uber from use of some of the disputed trade secrets/technology, which could be valuable for Google. Given that Uber hasn't monetized the technology yet, it would be hard to prove any damages. The outcome of a trial could have been a similar amount and an injunction and still been a big win for Google if the future self driving market is really going to be big and if the covered technology really was crucial.

No, Uber wasn't likely to settle for more, monetarily. The manner in which damages are fixed for trade secret and patent infringement make the total damages difficult to assess until the product goes to market. Google would be hard pressed to show actual loss, so Uber would likely be hit with some disgorgement stemming from unjust enrichment with some injunctive relief on the side.

What did they get? Unjust Enrichment sized damages and mutually agreed forbearance equivalent to injunctive relief.

And that's before we get into discussing how much of a minefield discovery was becoming in the case.

There is nothing in this settlement equivalent to injunctive relief. An injunction was the worst-case scenario for Uber and they would never settle for something that amounts to their worst-case, otherwise why not just finish the trial?

1) The most common and most valuable injunctive relief in IP litigation is an injunction to have people stop using your special sauce. That's literally what they got in the settlement.

2) Because prior to the trial there would have been a flurry of discovery related motions, a large amount of time pulling engineers off their jobs to have them perform witness prep at both companies, then a large period of pre-trial legal prep, then the trial itself.

At all stages, information is leaking, and expenses are going up.

... And then the appeals start.

... Then the partial re-trials and appeals on those.

It's a cost/benefit analysis. Waymo put forward a fairly weak case. There were damning elements of that case, for sure, but there were also weaknesses, and a case is a chain of reasoning that is only as strong as its weakest point. If Waymo was certain of victory they would not have settled, it would have been in their best interest to get things sorted in court and to get the maximum amount of money out of Uber. The fact that they settled is an indication that both parties were unsure of the outcome of the trial.

I think there might be still politics at play that GV is a investor at Uber. While Waymo is pretty unrelated to GV, there might have been "signals" from the parent company to quiet down the case.

Right. They extracted a fairly measly amount of money from Uber but more importantly they extracted a commitment that Uber would go the extra mile to prove they weren't making use of any of Waymo's IP.

I think the last point really drives is home. Following the trial for a week, there honestly wasn't any strong winner. It was pretty weak on both sides, and no one could've really predicted which side the jury would've fell I think. It was down to game theory, they would both rather get a settlement than bet on the jury.

Settlement isn't an indication that both parties were unsure of the outcome of the trial; trials are never certain things.

Settlement is the most common way civil litigation disputes are resolved. Trials are the anomaly.

In this case, both parties likely recognized that getting into a knife fight about documentary production in a high-stakes race-to-market matter was likely in Cruise's favor, not theirs.

8 trade secrets (down from 100s)

Was this by their own will? Back in the Oracle v Google trial, which had the same judge, he asked Oracle to cut down the 132 claims to only 3 - which Oracle was not happy about.

No. Judge Alsup told them to knock it down into the reasonable range. ("Less than 10") I believe it got stripped down to 9, and then one got thrown out of that group.

All the trade secrets claims were plan B for Waymo. The case was initially about patent infringement, and for undisclosed reasons they were all tossed out.

What was supposed to be the original slam dunk for Waymo, 'trade secret 96', Which referred to PCB circuit diagrams that were accidently emailed to Waymo instead of Uber ATG by a shared supplier, was even dropped. Waymo's whole case was pretty flimsy.

> The case was initially about patent infringement, and for undisclosed reasons they were all tossed out.

Possibly because the patents were invalid due to the existence of prior art? Previous discussion: https://news.ycombinator.com/item?id=15875685

The prior art cited in that news item only invalidated 'patent 936'. (That's my comment there, 3rd from top). The infringement claims were thrown out earlier than that (September 2017, IIRC).

>...The infringement claims were thrown out earlier than that (September 2017, IIRC).

According to the article, the infringement claims were dropped after Waymo was notified of the USPTO reexamination that was going to be done due to Swildens filing the ex parte reexamination request.

>...So he took the unusual step of challenging the 936 patent himself, filing what is called an ex-parte reexamination request. He gathered the prior art he had discovered, completed reams of paperwork, and pulled together his detailed arguments into a 101-page document that he filed with the USPTO on August 1.

Waymo dropped the claims relating to the 936 patent on Sep 13.

>...Swildens, naturally, thinks his reexamination request played a role. He notes that USPTO had a phone interview with Waymo’s lawyers a few days before, on September 8, at which the examiners presumably told Waymo that a reexamination of the 936 would be going ahead. If the reexamination started while the patent was being actively litigated, Uber and the court would have to be informed.

https://www.wired.com/story/eric-swildens-uber-waymo-lawsuit...

That had been a good long read; thanks for linking again

I asked about this back in the Oracle trial and was told that if you can't prevail on what you consider your top 10 allegations, you are not going to be able to prevail on the rest of them (assumption they are weaker than your top 10). As a result litigating all of them is not useful and only serves to confuse the jury.

I mean, yeah, 45 minutes is more than enough to exchange a ton of really critical information about a specific solution to a given technical/product problem (for example)...

    there isn’t a clear link to how those documents got onto Uber computers

Just a minor correction: Waymo has not demonstrated that the files went any further than the disks Levandowski had. To date, nothing belonging to Waymo has been found on Uber's servers.

I am not sure that is a minor correction. My guess is that Waymo bet on finding such evidence through discovery, but all it ended up with is armfuls of what is, at best, circumstantial evidence of collusion.

She makes a number of fair points, but:

> Can you really explain eight self-driving car trade secrets in 45 minutes?

Maybe I'm fooling myself, but I think I can do that in 45 seconds.

  - hey what's your learning rate scheduling for module X?
  - we use an EMA with coefficient 1e-4
  - oh great, and for Y, do you do constant strides?
  - no we use backtracking to figure out some sampling heuristic

etc.

None of which would be useful for a judge or jury. Heck, I'm generally pretty decent technically compared to the rest of the population and what you wrote here is barely relevant to me.

Sure what I just wrote is technobabble, I don't have much self-driving car expertise, but it was just to make a point. From my experience with large scale engineering projects, especially ones with ML involved, the tiniest details can make the stupidest difference.

Parameters are probably the lowest hanging fruit in any machine learning optimisation, being a trade secret is a big stretch.

Also use LBFG so you never worry about learning rate.

"Also use LBFG so you never worry about learning rate."

Literally never happens in practice.

Hyperparameter optimization for neural networks can be insanely expensive though[1], to the point where there are even startups (e.g: SigOpt) that do this. [1] https://twitter.com/beenwrekt/status/961263599674150912

the tweet you link to is architectural search, or figuring out what the neural network should look like (along with its parameters).

If self-driving car ML is "data scientists replacing a driver with computers", architecture search is computers replacing data scientists. $1.2M is roughly 3-4 data scientists' annual compensations at a company like Google or Uber.

Finding hyperparameters for an existing architecture is much easier because you would typically deal with a much lower cardinality in hyperparameter space. SigOpt is helping with that. Google has an internal solution that is able to take advantage of parameters within their job scheduling infrastructure (they actually have a bunch of them, the one I'm talking about is called hyperband [0]). CERN, as far as I know, was using hyperopt [1]. Avant uses an open-source version of SigOpt that we wrote in 2016 called loop [2].

[0] - https://arxiv.org/pdf/1603.06560.pdf [1] - http://hyperopt.github.io/hyperopt [2] - https://github.com/avantoss/loop

Before this gets inundated with people making comments about Googler data scientists making $400k, these are folks with PhDs in math or statistics, very accomplished academic literature, leaders in their fields. These are not people messing around with Haskell on the weekends and leaving an insurance firm's IT department to go to Google.

They also don't make $400k – an engineer that costs $400k doesn't get that in take home pay.

Entirely not true.

There are enough non-executive engineers at Google making half a million or more base/cash comp that you could fill a room with them. That's not the point.

Yea, and my point was that this can be expensive regardless of the semantics of whether or not architectural degrees of freedom are hyperparameters.

no one uses LBFG in deep learning or deep reinforcement learning, which is what self driving cars use

How long do you think it would take to explain that to 10 random strangers?

Not even ten random strangers. Ten strangers that are almost certainly not technical, since lawyers always try to dismiss smart people from juries?

As someone who has been on two juries, I object! Jury selection is complicated, and there are limitations to how many potential jurors can be struck. There are also multiple strategies that legal teams may adopt. In my last trial, the plaintiff made a very emotional case, but the defense presented a very logical case. There were two jurors who I would say were logically minded, and we were able to influence the rest of the jury, who were tending to stick on what I thought were irrelevant points. A lawyer friend of mine said that she has seen that one or two logically minded jurors can often steer the jury, and if she is able to put together a logical case, she wants to get some logical people on the jury. (However, being a public defender, she is not always able to put together a nice logical case. She has to do the best she can with what she has to work with.)

Aren't you proving cbhl's point? When she can't put together a logical case and has to rely on "this person has kids please don't send them to prison for this crime they obviously committed," she's going to be trying to keep off the most logical and articulate people at any costs.

chbl wrote: "Ten strangers that are almost certainly not technical, since lawyers always try to dismiss smart people from juries"

The point is that smart people are not always dismissed. Different legal strategies work better with different jury selections.

> When she can't put together a logical case and has to rely on "this person has kids please don't send them to prison for this crime they obviously committed," she's going to be trying to keep off the most logical and articulate people at any costs.

Nah, people who support nullification based on perception of the specific case and offender circumstances are not generally less logical and articulate than others.

You obviously want to eliminate people who either see the law as inherently just or see it as important to uphold the law even when the result would be unjust, as well as people inclined to see punishing your client as just despite the circumstances. But there's no reason, in general, to think that logical thinking or articulateness are, in general, going to be good correlated for those moral inclinations.

> lawyers always try to dismiss smart people from juries?

Lawyers don't “always try to dismiss smart people from juries”, or even try to identify smart people.

They do try to remove people that they expect will be more likely than average to be unfavorably inclined to their client or specific arguments, or that they expect to be a wildcard because they will be unable to restrict themselves to the courtroom presentation.

Neither of those, in general, is particularly correlated with intelligence.

The 45 minutes she's referring to, is the 45 minutes that Waymo spent talking about it in court. Not 45 minutes that Levandowski spent talking with Uber employees.

If all of the models line up a ton, an expert witness could explain that these the chances of reaching the same tuning on all these choices via independent work is practically 0.

So, I am sure this will get downvoted, but thi is not my field - but as a curious person.... can you please explain in-long-hand-but-still-like-I-am-5 what your comments means?

I dont understand it, but I want to.

coursera has a great course on self-driving cars. the only prerequisite is a college course in linear algebra and some python knowledge. If you don't have either of the two the course is still very much approachable - you'll just have to learn two (or three) things at the same time, so expect to double (or triple) the time you sink into it.

It's worth it

It is quite simple to explain. When your best guy gets hired away from your company, you are generally angry at the company that stole your employee, not the employee. Most of the tech is still trial and error so patents don't hold much water at this point unless you are TESLA. So this was always about "See what I can do to you because you hired my best guy"

Reply on news.ycombinator.com