The Battlefront Miscellaneous Thread

Rumors suggest Apple is developing custom silicon for AI servers. Given their scale and their cozy relationship with TSMC, I guess they'd be crazy not to cut out the middleman (NVIDIA) here.

It's not much of a suprise. Amazon, Meta, Microsoft, Tesla, and Google* already have their own in-house designed AI accelerators. The cost^ of training hardware is so enormous that even if you actually get on with Nvidia it may well still be worth designing your own chips. In fact the costs are so high that if you aren't designing chips you probably aren't capable of being at the leading edge of AI training/inference as you would potentially be wasting huge amounts of capital using off-the-shelf hardware.

Google for example added SparseCores (custom dataflow driven processors closely coupled to the memory controllers) to their TPUs to dramatically speed up their recommendation models. That kind of customisation can deliver large speed-ups over more generic AI hardware, if you operate at the scale that makes it worthwhile.

You might ask: why buy Nvidia GPUs if custom hardware is potentially a lot better? Because external customers want Nvidia GPUs and CUDA, they don't have the luxury of designing hardware for their own problem, they want the "cheapest" standard AI hardware.

* Google is on their 6th/7th generation of TPU now, and it's not just the TPUs, they are also building custom networking hardware and optical switches as well in order to scale-up.

^ ~$10 billion worth of H100 and other Nvidia GPUs for Meta. Microsoft and OpenAI are looking at a $100 billion investment for their Stargate AI supercomputer.
 

ZnU

Ars Legatus Legionis
11,694
Znu, I'm not sure why you think share of TSMC's line helps here? Because in fact, they would require MORE than their current 25% to accomplish anything here. Certainly they aren't going to reduce the number of iPhones being produced to ramp up on AI chips.

You seemed to think it was significant that NVIDIA was "only a chipmaker." The fact that Apple nonetheless makes more chips seems relevant. They also make more chips than AMD or maybe even Intel (hard to nail down exact numbers). I don't think people have quite internalized how serious Apple's chip business now is.

Capacity isn't fixed over time, and Apple has shown a past willingness to e.g. help TSMC fund process advances in exchange for preferential access.

I doubt TSMC has any additional capacity to give, they're already building AI chips for companies that are ahead of Apple by significant margins.

Cutting out the middleman, Apple could outbid NVIDIA, effectively splitting NVIDIA's margin with TSMC. This is a standard benefit of vertical integration, and helps explain why pretty much everyone in a position to be a major customer for their own AI hardware — Microsoft, Amazon, Meta, Google — has or is planning custom silicon.

If Intel wanted to, you know, get their act together on the foundry side any time here, there would be appetite for their lines. They need a process node that doesn't suck.

Yeah, it's a little embarrassing that Intel is about to ship what appears to be a perfectly respectable AI accelerator... made on TSMC 5nm.
 
If Intel wanted to, you know, get their act together on the foundry side any time here, there would be appetite for their lines. They need a process node that doesn't suck.

The same holds for any attempt by them at an AI chip. But I won't hold my breath.
Intel has really fallen. They just seem to be missing everything.
 

wco81

Ars Legatus Legionis
28,661
Lately, there's been more discussion about how technology is making people unhappy, specifically teens.

One of the people making this argument is Jonathan Haidt, an NYU Stern Business School professor where he deals in the areas of ethics but is described as a social psychologist.

Haidt has been making the rounds on various media outlets, talking about how starting around 2010-2011, teens started becoming actively harmed, prone to depression and self-harm ideation, because of social media.

He says that their brains are being rewired to social network app notifications, which make them addictive to content which is harmful for their mental health.

He calls out iPhones but he's referring to all smart phones and he specifically focuses on social networking apps. He says he tells his students to turn off notifications on their phones and supposedly their moods have improved.

Oh he's written at least one best-selling book on this topic, so he has an incentive to raise the issue in the media. He advocates that children not be given phones until they're at least 16. His argument is that in adolescence the brain pathways haven't had a chance to be set so it makes children younger than their mid teens particularly vulnerable to the worst effects of the dopamine-reward mechanism triggered by social network app notifications.

One of the biggest critiques of Haidt's hypothesis was published recently in Nature:

Two things need to be said after reading The Anxious Generation. First, this book is going to sell a lot of copies, because Jonathan Haidt is telling a scary story about children’s development that many parents are primed to believe. Second, the book’s repeated suggestion that digital technologies are rewiring our children’s brains and causing an epidemic of mental illness is not supported by science. Worse, the bold proposal that social media is to blame might distract us from effectively responding to the real causes of the current mental-health crisis in young people.

Haidt asserts that the great rewiring of children’s brains has taken place by “designing a firehose of addictive content that entered through kids’ eyes and ears”. And that “by displacing physical play and in-person socializing, these companies have rewired childhood and changed human development on an almost unimaginable scale”. Such serious claims require serious evidence.




Collection: Promoting youth mental health

Haidt supplies graphs throughout the book showing that digital-technology use and adolescent mental-health problems are rising together. On the first day of the graduate statistics class I teach, I draw similar lines on a board that seem to connect two disparate phenomena, and ask the students what they think is happening. Within minutes, the students usually begin telling elaborate stories about how the two phenomena are related, even describing how one could cause the other. The plots presented throughout this book will be useful in teaching my students the fundamentals of causal inference, and how to avoid making up stories by simply looking at trend lines.



My question is whether Haidt's timeline even makes sense. iPhone shipped in 2007. Then we had the Great Financial Crisis. People were losing jobs and homes.

Was there a lot of smart phones being handed down to kids in 2010 and 2011?

Or did that become more of a thing in the mid 2010s and later?

Smart phone sales did take off during this period but global unit sales didn't top 500 million until 2012 and didn't top 1 billion until 2014.


YEARSMARTPHONES
(Millions of units)
NON-SMARTPHONES / FEATURE PHONES
(Millions of units)
TOTAL MOBILE PHONES
(Millions of units)
2009172.38994.081166.46
2010296.651296.821593.47
20114721302.851774.85
2012680.111110.131790.24
2013969.72846.681816.4
20141219.3635.21854.5
20151314449.51763.5
20161442.7371.21813.9
20171479.1324.31803.4
20181522298.41820.4
20191532284.81816.8
20201397.4251.71649.1
20211510.3237.51747.8
20221516.4233.31749.7
20231594228.31822.3
 

Schpyder

Ars Tribunus Angusticlavius
9,692
Subscriptor++
If anything is affecting the mental health and happiness of teens, I'd say it's probably a combination of helicopter parents, the loss of third spaces, the loss of free/leisure time due to increasingly demanding homework and extracurriculars, greater social awareness of existential threats from climate change and capitalism run amok, and the general economic malaise that has existed since the great recession. Doomscrolling and toxic social media might exacerbate the problem, but pinning it primarily on that has always seemed an incredible stretch to me.
 
He's not the only one saying this and I can't vouch for the timeline, but I think the other side of the coin is social media which prioritizes anger and disagreement along with making bullying even easier.

Twitter and Tik Tok seem to come up most often in this discussion with Twitter being the worse of the 2, basically, to be successful on twitter, you need to be highly combative. It leads to all those negatives.
 

lithven

Ars Tribunus Militum
1,932
basically, to be successful on twitter, you need to be highly combative. It leads to all those negatives.
Do people try to be "successful" on social media? I'm at a loss for what that even mean. I know there are "influencers" that can make some money on various social media platforms by hocking various products of dubious value but, for the most part I don't see diving head first into the rage machine being the best way to break into such an endeavor.

I guess "going viral" is some sort of success but I've seen enough of that to know that isn't always the best thing for anyone. I seem to recall instances that ultimately resulted in the affected person deleting their social media presence completely (but to be fair I honestly haven't even seen a small fraction of the stuff that supposedly goes viral).
 

Nevarre

Ars Legatus Legionis
24,110
Do people try to be "successful" on social media? I'm at a loss for what that even mean. I know there are "influencers" that can make some money on various social media platforms by hocking various products of dubious value but, for the most part I don't see diving head first into the rage machine being the best way to break into such an endeavor.

I guess "going viral" is some sort of success but I've seen enough of that to know that isn't always the best thing for anyone. I seem to recall instances that ultimately resulted in the affected person deleting their social media presence completely (but to be fair I honestly haven't even seen a small fraction of the stuff that supposedly goes viral).

Absolutely yes. It's like a get rich quick scheme that works out for just enough people to keep hope alive in every other teenager/young person. Obviously not everyone wants to follow the "influencer" path, but it's the new celebrity and there's just enough belief that it could happen to anyone, anywhere, as long as they put out enough good content. (read: LOTS and LOTS of content--and you can never ever fail or you get that hate response back and that isn't healthy at all.)

One of my kid's friends started as the "popular" kid in school and got a few thousand TikTok followers as a young teenager. They're now about to graduate high school with >1M followers and moving with their boyfriend (who also has a 7-figure TikTok follower base) to California to pursue that dream of at least continuing on the trajectory as a social media star if nothing else. My kid's path is not heading towards "influencer" status but she would if she could and is as they say -- lowkey jealous of the friend.

If you get big enough, there's money in it. Only the 1% of the 1% make big money, but it's enough to keep that hope alive, just like acting, music, professional sports, etc.
 
If anything is affecting the mental health and happiness of teens, I'd say it's probably a combination of helicopter parents, the loss of third spaces, the loss of free/leisure time due to increasingly demanding homework and extracurriculars, greater social awareness of existential threats from climate change and capitalism run amok, and the general economic malaise that has existed since the great recession. Doomscrolling and toxic social media might exacerbate the problem, but pinning it primarily on that has always seemed an incredible stretch to me.
Add school shootings for kids in US.

The amount of homework is bad though.
 

wco81

Ars Legatus Legionis
28,661
Where does OpenAI get revenues?

From subscriptions to ChatGPT right?

So why would they pay Apple? Is it to be the default LLM or something?

The Google deal is under heavy scrutiny so how would they cut a deal with OpenAI or Google without attracting heat?


It appears OpenAI could be liable for all kinds of copyright violations. They just ingested any and every content they could with impunity and their defense is going to be fair use. If the courts don't buy that, they not only may not be able to pay but any company which distributes ChatGPT may also have liability.
 

Horatio

Ars Legatus Legionis
24,069
Moderator
Where does OpenAI get revenues?

From subscriptions to ChatGPT right?
This wouldn't provide the kind of money Apple would be asking for. MS would have to front it.
So why would they pay Apple? Is it to be the default LLM or something?
Yup, just like Google paid to own the search box, this deal would be to own the LLM (possibly both off-device, and to provide an on-device model), consider, for example, a Siri replacement.
The Google deal is under heavy scrutiny so how would they cut a deal with OpenAI or Google without attracting heat?
If they go with Google, it would increase the heat yes, but probably not if they went with OpenAI, especially because OpenAI is not and will not be the default LLM on Android devices, which is a significant departure from the analogy to the search box.
It appears OpenAI could be liable for all kinds of copyright violations. They just ingested any and every content they could with impunity and their defense is going to be fair use. If the courts don't buy that, they not only may not be able to pay but any company which distributes ChatGPT may also have liability.
Presumably Apple would insist on indemnity. MS provides LLM indemnity, so there is precedent.
 
  • Like
Reactions: Nevarre

Shavano

Ars Legatus Legionis
59,253
Subscriptor
You seemed to think it was significant that NVIDIA was "only a chipmaker." The fact that Apple nonetheless makes more chips seems relevant. They also make more chips than AMD or maybe even Intel (hard to nail down exact numbers). I don't think people have quite internalized how serious Apple's chip business now is.

Capacity isn't fixed over time, and Apple has shown a past willingness to e.g. help TSMC fund process advances in exchange for preferential access.

Cutting out the middleman, Apple could outbid NVIDIA, effectively splitting NVIDIA's margin with TSMC. This is a standard benefit of vertical integration, and helps explain why pretty much everyone in a position to be a major customer for their own AI hardware — Microsoft, Amazon, Meta, Google — has or is planning custom silicon.
TSMC gets paid the same either way, since neither NVIDIA nor Apple makes chips.
Yeah, it's a little embarrassing that Intel is about to ship what appears to be a perfectly respectable AI accelerator... made on TSMC 5nm.
That's what the US government is so concerned about. Almost all of our high end chip vendors depend on ONE country that China has designs on controlling. They're throwing billions at the problem. Possibly not enough billions to really make a change.
 

ZnU

Ars Legatus Legionis
11,694
That's what the US government is so concerned about. Almost all of our high end chip vendors depend on ONE country that China has designs on controlling. They're throwing billions at the problem. Possibly not enough billions to really make a change.

Intel claims it will be back on top later this year with its 18A process. Volume is always the tricky part, though. We'll see.

Even if the US screws up its attempts to onshore chip production (quite possible; my default assumption is that US government spending mostly gets squandered on regulatory nonsense and/or corporate graft), Japan and South Korea are also investing heavily, and are better at making physical stuff these days. One way or another, Taiwan likely won't be so critical in four or five years.

I think China will wait for this to play out. Not only because the US will be less motivated to put up a fight at that point, but also because the balance of military power is just generally shifting in China's favor over time, as it reaps the benefits of a larger industrial base.
 

Chris FOM

Ars Legatus Legionis
10,001
Subscriptor
China has roughly 10 years, tops, before the incoming demographic collapse materializes and they are forced to turn inwards. Doesn’t mean they can’t cause a ton of damage between now and then but time doesn’t favor them at all, they’re facing a rapidly approaching deadline before they’re massively diminished as a major world power.
 
We're moving into soapbox territory, but the current War in Ukraine is showing just how bad the offshoring on Manufacturing was for the US. We simply can't produce enough war time technology on home soil anymore.

Having said that, Korea and Japan would be perfectly fine for US interests. Friend-shoring is the concept.

The default assumption about government spending getting squandered show just how ingrained Reagan era bias has become. We as a country can't even remember that in fact, the US government was actually good at this stuff before then.

I also think it's as much State and local regulation.

For Intel as a specific example, they need competant management AND the funding to spend money freely. The money alone isn't enough.

I hope for the general good of the industry that Intel figures this out...Or someone else steps up. But it's not clear now if they will/can.
 

ZnU

Ars Legatus Legionis
11,694
China has roughly 10 years, tops, before the incoming demographic collapse materializes and they are forced to turn inwards. Doesn’t mean they can’t cause a ton of damage between now and then but time doesn’t favor them at all, they’re facing a rapidly approaching deadline before they’re massively diminished as a major world power.

Sure, but that just bookends things from the other side. The late '20s into the early '30s are the most likely time for them to make a move.
 
Last edited:
Where does OpenAI get revenues?

From subscriptions to ChatGPT right?

So why would they pay Apple? Is it to be the default LLM or something?

The Google deal is under heavy scrutiny so how would they cut a deal with OpenAI or Google without attracting heat?


It appears OpenAI could be liable for all kinds of copyright violations. They just ingested any and every content they could with impunity and their defense is going to be fair use. If the courts don't buy that, they not only may not be able to pay but any company which distributes ChatGPT may also have liability.
I guess what I don't understand is how they would make money by being on iPhone. How does the inputs help? Seems like Apple should be paying to use their tech.
 

Schpyder

Ars Tribunus Angusticlavius
9,692
Subscriptor++
I guess what I don't understand is how they would make money by being on iPhone. How does the inputs help? Seems like Apple should be paying to use their tech.

Oh, they're not going to, but no one cares. Stock price doesn't have anything to do with actual revenues for new tech, it's all vibes. As long as the outlook for growth is significant, then the share price goes up, up, up! And saying you landed a major client like Apple will massively increase their mindshare (even if they have to pay for the privilege), which means stock price up, which means investors happy.

No one gives a shit about the long-term and/or late-coming rube investors who are going to eventually get left holding the bag after every other smaller idiot has cashed out.
 
Oh, they're not going to, but no one cares. Stock price doesn't have anything to do with actual revenues for new tech, it's all vibes. As long as the outlook for growth is significant, then the share price goes up, up, up! And saying you landed a major client like Apple will massively increase their mindshare (even if they have to pay for the privilege), which means stock price up, which means investors happy.

No one gives a shit about the long-term and/or late-coming rube investors who are going to eventually get left holding the bag after every other smaller idiot has cashed out.
So you are saying that it isn't about profit or loss for the AI companies, but share price?
 

JimCampbell

Ars Tribunus Militum
2,412
Subscriptor
So you are saying that it isn't about profit or loss for the AI companies, but share price?
Right now, this is just another bubble waiting to burst. I spent sometime recently chatting to an old friend who’s a professor in a well-regarded computer science department at a UK university and whose speciality is in machine learning.

He was very clear in his belief that both the current capability and potential advances in the short to medium term are being vastly overstated.

(I’m not saying there’s no value there… but I think most of us here are old enough to remember multiple instances of the New Hotness in tech, the flurry of massively over-valued start-ups, the inevitable market ‘correction’ and the eventual emergence of something useful, once the hype dies down.)
 

Schpyder

Ars Tribunus Angusticlavius
9,692
Subscriptor++
So you are saying that it isn't about profit or loss for the AI companies, but share price?

It's all about share price and profit/loss are largely irrelevant for any hyped-up new-hotness companies. There comes a tipping point where they're no longer in vogue and have to deal with being Mature Companies rather than coasting on startup hype, but that can take a loooooong time to happen. Tesla has only finally come down to earth a bit over the last couple of years, now that the EV hype train has derailed a bit. There was a little bit of a correction from investors following Uber's disastrous IPO in 2019, but sentiment quickly shifted again to FOMO and chasing unicorns.
 
It's all about share price and profit/loss are largely irrelevant for any hyped-up new-hotness companies. There comes a tipping point where they're no longer in vogue and have to deal with being Mature Companies rather than coasting on startup hype, but that can take a loooooong time to happen. Tesla has only finally come down to earth a bit over the last couple of years, now that the EV hype train has derailed a bit. There was a little bit of a correction from investors following Uber's disastrous IPO in 2019, but sentiment quickly shifted again to FOMO and chasing unicorns.
So...Google, MS, Meta are not mature companies?

Especially MS...It seems that Satya et al. are not as gung-ho on losing tons of money for payouts a decade from now. (which is ironic in that all the big stuff going on in MS is from those kinds of 10-year out bets).
 

cogwheel

Ars Tribunus Angusticlavius
6,691
Subscriptor
You seemed to think it was significant that NVIDIA was "only a chipmaker." The fact that Apple nonetheless makes more chips seems relevant. They also make more chips than AMD or maybe even Intel (hard to nail down exact numbers). I don't think people have quite internalized how serious Apple's chip business now is.
Does Apple just make more chips, or do they also make more chip (i.e. area)? The latter should be the important part for TSMC, since the only thing pulling in the opposite direction is packaging.
  • The M3 is about the same size as the AD107 (the chip in the RTX 4060), Nvidia's smallest of this generation
  • The M3 Pro looks to fit in between AD106 (RTX 4060 Ti) and AD104 (RTX 4070/S/Ti) but much closer to AD106 in size
  • The M3 Max might be as large as the AD103 (RTX 4070TiS, RTX 4080/S) but may fit in between AD104 and AD103
  • The A17 is smaller than anything Nvidia makes, and is Apple's volume leader by far
  • Apple has nothing as gargantuan as the AD102 (RTX 4090), which is about the size of six A17s put together, or the even crazier non-consumer stuff like the H100 (eight A17s)
  • The Apple R1 (unknown size, not even an estimate) is too low volume to matter even if it's the size of the stuff Cerebras makes (which it obviously isn't)
  • The rest of the stuff Apple makes is both much smaller than the A17 and much lower volume.
(Nvidia also uses all of the AD-series chips in their business and enterprise-targeted cards in addition to the consumer cards I've named)

Given that Apple almost certainly has a significant overall die count lead, is it enough to overcome Nvidia's larger average die size, and if so, by how much? My gut feeling is Apple does ship more mm2 of chip than Nvidia, but I expect the margin to be much smaller than the raw chip count would imply, and also wouldn't be surprised if Nvidia edges out the lead.
 

Mark086

Ars Tribunus Angusticlavius
10,595
They just ingested any and every content they could with impunity and their defense is going to be fair use.

And why wouldn't fair use be sufficient?

Do we withdraw every degree issued in the past decade to any students that used pirated materials to study from?

I'm all for it, go back as many decades as you want; but I don't think that actual serves anyone's interest.
 
Last edited:
And why wouldn't fair use be sufficient?

Do we withdraw every degree issued in the past decade to any students that used pirated materials to study from?

I'm all for it, go back as many decades as you want; but I don't think that actual serves anyone's interest.

To study from — or to quote and reproduce content for their papers from for academic use? These have always been a question of going to the library, requesting a source from another library elsewhere, or pulling the information off the microfiche reader.

But to offer to the world at large, as a commercial business, for entertainment purposes?

Come on. Please don't try to argue that these things are the same.
 
No, but they are afraid of the markets punishing them for not being seen as moving quickly enough on the New Tech Hotness™ — if that impacts their share price, then the senior execs will get serious flak from the shareholders.
Kind of like Meta losing billions on the Meta verse?

I guess I am having a hard time believing that MS or Google would be willing to spend billions of dollars to be on iPhone with the sole hope of return as "share price goes up". There has to be something they get out of it besides that. User data? Inputs? something more valuable than "halo affect" (which if anyone will sign up for that it would be MS :) )
 
Last edited:
To study from — or to quote and reproduce content for their papers from for academic use? These have always been a question of going to the library, requesting a source from another library elsewhere, or pulling the information off the microfiche reader.

But to offer to the world at large, as a commercial business, for entertainment purposes?

Come on. Please don't try to argue that these things are the same.
I think you might be find other examples. Books for example.
 
People use other works, quote them and such "to the world at large, as a commercial business, for entertainment purposes?" It would fall under fair use for that, why not for LLM?

If you create a literary work entirely derivative of existing work — and recognisably so — see how far you get once you're successful enough to be a target worth suing.
 

ZnU

Ars Legatus Legionis
11,694
Given that Apple almost certainly has a significant overall die count lead, is it enough to overcome Nvidia's larger average die size, and if so, by how much? My gut feeling is Apple does ship more mm2 of chip than Nvidia, but I expect the margin to be much smaller than the raw chip count would imply, and also wouldn't be surprised if Nvidia edges out the lead.

The 25% vs. 11% is by revenue, not die count. So it should (literally) price a lot of this in.
 

cogwheel

Ars Tribunus Angusticlavius
6,691
Subscriptor
And why wouldn't fair use be sufficient?

Do we withdraw every degree issued in the past decade to any students that used pirated materials to study from?

I'm all for it, go back as many decades as you want; but I don't think that actual serves anyone's interest.
The problem, which @analogika is trying to get at, isn't the ingestion per se, it's the regurgitation. The NYT is suing OpenAI not because ChatGPT reads the NYT, but because it is, or was, pretty easy to get ChatGPT to reproduce entire or large portions of NYT articles.

ChatGPT "reading" something shouldn't be a problem by itself, but ChatGPT doesn't read like humans. It doesn't understand facts, merely word presence and proximity, so it doesn't form "memories" anything like humans do. When it heaves an answer at you, it doesn't really summarize a concept (since it doesn't even have the concept of a concept), so it can only give you what someone else wrote that it "read", or a word salad made of the writings of multiple people. ChatGPT is good enough now that when it makes salad, it's at least palatable most of the time, but that doesn't change the fact that the ingredients are all sourced from humans, and OpenAI has undeniably admitted that they have fed ChatGPT copyrighted works which get re-served to you in whole or in part (and sometimes large part) for commercial reasons.

People use other works, quote them and such "to the world at large, as a commercial business, for entertainment purposes?" It would fall under fair use for that, why not for LLM?
Have you actually looked at what uses are considered fair use, and the fair use test criteria?

Fair use is for "commentary, search engines, criticism, parody, news reporting, research, and scholarship" (quote from Wikipedia), and ChatGPT is not marketed as primarily a tool for any of those. The only one I can see anyone trying to weasel into is "news reporting", but when ChatGPT barfs out copyrighted news articles (or large parts of them), the articles themselves are not the news so it still very clearly fails.

The fair use test, which has generated that list of generally accepted forms of use above, has four parts (quoting from 17 U.S. Code § 107):
  1. the purpose and character of the use, including whether such use is of a commercial nature or is for nonprofit educational purposes;
  2. the nature of the copyrighted work;
  3. the amount and substantiality of the portion used in relation to the copyrighted work as a whole; and
  4. the effect of the use upon the potential market for or value of the copyrighted work.
ChatGPT would obviously have issues under parts one (it's for commercial use), three (it can be made to vomit up large chunks of copyrighted works), and four (when ChatGPT disgorges that copyrighted content, OpenAI may get paid, but the copyright owner does not at this point).

The 25% vs. 11% is by revenue, not die count. So it should (literally) price a lot of this in.
I missed your earlier post that mentioned this. I agree, it should largely reflect chip area volume, with only the fact that Apple is, on average, on a newer process than Nvidia, but that shouldn't be enough to counteract such a large difference.
 
Fair use is for "commentary, search engines, criticism, parody, news reporting, research, and scholarship"
Oh...I could see them arguing that it is basically a search engine. I mean it takes it all in and then assembles it to provide the results. I could also imagine them saying research, but search is a better argument in my mind.