It's pretty nervy for Chhabria to accuse Alsup of "blowing off the most important factor" when they see each other in the courthouse lunchroom every day.
I agree that Chhabria appears to assume a lot of facts that are only in evidence in the minds of certain people on the S.F. peninsula, not so much in the real world.
Their press releases all say things like they grew by 65% without any actual numbers, which tells us they're pretty small so $1M could be painful.
I used them for VoIP service a while ago. They were meh.
I agree that this decision is absurd, but Cognac is indeed a GI. If you want to call something Cognac, it has to be made in the region around the commune of Cognac, France, and also meet other rules about the kinds of grapes and the way they're processed.
In 2006 the Field vs. Google case clarified the legality of Google's search engine cache. The judge found that the cache was legal, partly as fair use but also because it was easy to opt out using robots.txt. If crawlers are ignoring the robots file, the analysis changes and not in a way that is good for the crawlers.
https://en.wikipedia.org/wiki/Field_v._Google%2C_Inc.
I was Google's technical expert in the case.
In the UK, the government does not provide number plates. You buy them from private vendors. The vendors are supposed to verify that you own the vehicle to which the number is assigned, but I expect at the right price the annoying documentation can be waived.
Andrew Odlyzko has been writing about micropayments ("they're like gallium arsenide, the technology of the future and always will be") for two decades, arriving at conclusions similar to yours.
He points out that phone companies had this debate a century ago, metered vs. flat rate local calls. In the US with private phone companies, they all went to flat rate, in Europe with PTTs they were until recently all metered.
https://www-users.cse.umn.edu/~odlyzko/doc/case.against.micropayments.pdf
I'm a high status Aeroplan member, and he's right. It is a pain to use Air Canada points to book flights, and it is double pain to figure out the best combination of points you have, certificates that change the number of points needed, and purchased points if you're close to having enough. That is surely not an accident, since every seat booked with points is one they can't sell.
I'm not surprised that they would be annoyed at someone who figures that out for you.
LLMs do indeed give you a token string, but it's what comes out that matters. You probably remember plagiarism court cases where a composer did not remember listening to the song he copied, but it was clear he did so he lost. Similarly, it doesn't matter how Stability is generating Getty trademarks, just that they were in the training data and they show up in the output.
I expect that now that the LLM crowd is realizing that it's a problem that they can't trace input to output, they'll at least try to fix it.
I have written a lot of books and back in the day made a lot of money from them. I agree that it is fine for AI models to chop them up and use them for training data. The Google Books case should have settled that, since one of the reasons the court found for Google is that they used the scanned books to produce things unlike copies of books, an index with snippets, interesting research like tracking the usage of words over the years.
The problem is that LLMs cannot tell you where their results come from, and if they start regurgitating recognizable chunks of my books, I would not be happy. (I would be differently unhappy if they invented nonsense quotes and claimed I wrote them.)
The Getty Images suit against Stability AI shows examples of generated images with recognizable Getty watermarks, so I have a lot of sympathy for Getty there. If LLMs could show their work, so we could tell the difference between research and cut'n'paste, these arguments would be a lot easier to resolve.
I am much more worried about the music publishers' suit, where IA is much more reckless than they were with CDL. In the CDL suit, they at least had a theory about why it was legal. With the 78's, their argument appears to be that the 100 year term on recordings is stupid, which it is, but that's not going to impress a judge. There is a process to show that a recording is abandoned which IA hasn't even tried to use (it's tedious but not all that hard) and that's just going to annoy a judge.
The book publishers could have asked for damages but didn't, because they have business reasons not to damage IA. But the music industry has always hated the Internet and they're not going to be as nice when (not if) they win.
FWIW even though Tucows' head office is in Toronto, it's a US company, incorporated in Pennsylvania.
I suspect that if you asked, they'd tell you that the legal environment in Canada makes it impossible to compete with the entrenched oligopoly of Bell, Rogers, and Telus. The CRTC is too busy with stuff like shaking down Facebook and Google with link taxes.
How about assuming that Elon ate another bad musuhroom or something, and that his reply had nothing to do with the putative topic? As the article carefully determined, we know there's no connection based on coherence or logic.
While I have no moral sympathy for the record labels, this suit was 100% guaranteed when the MMA gave recordings 100 years of copyright. It's stupid, but it's the law and judges enforce the law.
IA did itself no favors by poking the record companies in the eye since everyone knows how aggressive they are. The MMA provides a painful but not impossible way to get access to orphan works, do a lot of searches to see if it's still in print anywhere, then if you don't find anything send a spreadsheet of them to the Copyright Office, and if nobody objects in 90 days you can use it. I expect most of the recordings from the 1920s and 30s are orphaned, but IA hasn't even tried to do this. They could probably get tens of thousands of records cleared and leave to the RIAA the stuff you can get on Spotify.
I presume Salesforce will ask for en banc rehearing, and I would hope there's a reasonable chance since this decision is so far in the weeds compared to other circuit decisions.
I've done a lot of mobile deposits and I do not ever remember the images staying on the phone. They're uploaded to the bank.
I suppose it might be useful to find that there was a bank app on the phone set up to log into the suspect's account, but that's not what they said they're looking for.
The real question
is whether Noem is smart enough to understand how stupid the stuff she is saying is. My impression is that she isn't.
blow me off
It's pretty nervy for Chhabria to accuse Alsup of "blowing off the most important factor" when they see each other in the courthouse lunchroom every day. I agree that Chhabria appears to assume a lot of facts that are only in evidence in the minds of certain people on the S.F. peninsula, not so much in the real world.
Multiple Trumps
But there are multiple Trumps. One's named Don, the other's Elon. Glad I could clear that up.
How big is Lingo?
Their press releases all say things like they grew by 65% without any actual numbers, which tells us they're pretty small so $1M could be painful. I used them for VoIP service a while ago. They were meh.
breath mint and floor wax
I agree that this decision is absurd, but Cognac is indeed a GI. If you want to call something Cognac, it has to be made in the region around the commune of Cognac, France, and also meet other rules about the kinds of grapes and the way they're processed.
There is case law
In 2006 the Field vs. Google case clarified the legality of Google's search engine cache. The judge found that the cache was legal, partly as fair use but also because it was easy to opt out using robots.txt. If crawlers are ignoring the robots file, the analysis changes and not in a way that is good for the crawlers. https://en.wikipedia.org/wiki/Field_v._Google%2C_Inc. I was Google's technical expert in the case.
This is a lot easier in the UK
In the UK, the government does not provide number plates. You buy them from private vendors. The vendors are supposed to verify that you own the vehicle to which the number is assigned, but I expect at the right price the annoying documentation can be waived.
This is a very old discussion
Andrew Odlyzko has been writing about micropayments ("they're like gallium arsenide, the technology of the future and always will be") for two decades, arriving at conclusions similar to yours. He points out that phone companies had this debate a century ago, metered vs. flat rate local calls. In the US with private phone companies, they all went to flat rate, in Europe with PTTs they were until recently all metered. https://www-users.cse.umn.edu/~odlyzko/doc/case.against.micropayments.pdf
The point of points is not to redeem them
I'm a high status Aeroplan member, and he's right. It is a pain to use Air Canada points to book flights, and it is double pain to figure out the best combination of points you have, certificates that change the number of points needed, and purchased points if you're close to having enough. That is surely not an accident, since every seat booked with points is one they can't sell. I'm not surprised that they would be annoyed at someone who figures that out for you.
classes are not AI
The usual reason is that the students are reading copies of the book that they bought, not ones that they stole.
Regurgiwhatever
LLMs do indeed give you a token string, but it's what comes out that matters. You probably remember plagiarism court cases where a composer did not remember listening to the song he copied, but it was clear he did so he lost. Similarly, it doesn't matter how Stability is generating Getty trademarks, just that they were in the training data and they show up in the output. I expect that now that the LLM crowd is realizing that it's a problem that they can't trace input to output, they'll at least try to fix it.
This is very fact specific
I have written a lot of books and back in the day made a lot of money from them. I agree that it is fine for AI models to chop them up and use them for training data. The Google Books case should have settled that, since one of the reasons the court found for Google is that they used the scanned books to produce things unlike copies of books, an index with snippets, interesting research like tracking the usage of words over the years. The problem is that LLMs cannot tell you where their results come from, and if they start regurgitating recognizable chunks of my books, I would not be happy. (I would be differently unhappy if they invented nonsense quotes and claimed I wrote them.) The Getty Images suit against Stability AI shows examples of generated images with recognizable Getty watermarks, so I have a lot of sympathy for Getty there. If LLMs could show their work, so we could tell the difference between research and cut'n'paste, these arguments would be a lot easier to resolve.
minor error
In several places, New York is in the Second circuit, not the Sixth. I live near Elmira and wow, I had no idea.
Speaking of wishful thinking
I am much more worried about the music publishers' suit, where IA is much more reckless than they were with CDL. In the CDL suit, they at least had a theory about why it was legal. With the 78's, their argument appears to be that the 100 year term on recordings is stupid, which it is, but that's not going to impress a judge. There is a process to show that a recording is abandoned which IA hasn't even tried to use (it's tedious but not all that hard) and that's just going to annoy a judge. The book publishers could have asked for damages but didn't, because they have business reasons not to damage IA. But the music industry has always hated the Internet and they're not going to be as nice when (not if) they win.
Who'd have expected
Judge Cannon wrote a well reasoned decision that gets the law and the merits right. Perhaps there's hope for her after all.
Canadian?
FWIW even though Tucows' head office is in Toronto, it's a US company, incorporated in Pennsylvania. I suspect that if you asked, they'd tell you that the legal environment in Canada makes it impossible to compete with the entrenched oligopoly of Bell, Rogers, and Telus. The CRTC is too busy with stuff like shaking down Facebook and Google with link taxes.
Null hypothesis
How about assuming that Elon ate another bad musuhroom or something, and that his reply had nothing to do with the putative topic? As the article carefully determined, we know there's no connection based on coherence or logic.
What the law should be isn't what the law is
While I have no moral sympathy for the record labels, this suit was 100% guaranteed when the MMA gave recordings 100 years of copyright. It's stupid, but it's the law and judges enforce the law. IA did itself no favors by poking the record companies in the eye since everyone knows how aggressive they are. The MMA provides a painful but not impossible way to get access to orphan works, do a lot of searches to see if it's still in print anywhere, then if you don't find anything send a spreadsheet of them to the Copyright Office, and if nobody objects in 90 days you can use it. I expect most of the recordings from the 1920s and 30s are orphaned, but IA hasn't even tried to do this. They could probably get tens of thousands of records cleared and leave to the RIAA the stuff you can get on Spotify.
En banc?
I presume Salesforce will ask for en banc rehearing, and I would hope there's a reasonable chance since this decision is so far in the weeds compared to other circuit decisions.
Check images?
I've done a lot of mobile deposits and I do not ever remember the images staying on the phone. They're uploaded to the bank. I suppose it might be useful to find that there was a bank app on the phone set up to log into the suspect's account, but that's not what they said they're looking for.