Re: Yeah, but
"Who can tell the difference between a genuine video of the big Cheeto blabbering and a deep fake?"
Having trouble telling a deep fake from the real thing?
That's a job for Fred Sassy - https://www.youtube.com/watch?v=9WfZuNceFDM
629 publicly visible posts • joined 15 Jun 2010
Given that this entire swathe of companies has been enabled by primary legislation from Congress, maybe they should look at changing the law they wrote that allows social media to act like publishers without any of the liabilities or responsibilities publishers have?
|IF, and that's a big "if", the AI is transformative enough, creating something that is original but inspired by another work, is that any different than a human?
https://www.copyright.gov/comp3/chap300/ch300-copyrightable-authorship.pdf
Section 308.
Whatever is created by an AI can only be purely derivative, as under US copyright law it would not qualify for copyright itself under section 308 , it doesn't contain any creativity as defined in (US) law.
"Which is a simplistic statement and, as given, is also true for every one of the "fair use" exemptions - satire, critical review etc."
ChatGPT does not restrict it's output to such uses. In any case the derivative work happens much earlier in the process, when the model is updated as a result of being trained on the work in question. The secondary derivatives are largely irrelevant except to show that the model itself is clearly derivative of its inputs
" Not sure that "spark of creativity" has a legal meaning?"
Literally the basis of copyright protection in the US
https://www.copyright.gov/what-is-copyright/
Copyright is originality and fixation
Original Works
Works are original when they are independently created by a human author and have a minimal degree of creativity. Independent creation simply means that you create it yourself, without copying. The Supreme Court has said that, to be creative, a work must have a “spark” and “modicum” of creativity.
"And the "netflux" logo example is showing decent Fair Use, btw. So in that case, it was all working ok."
In what context is this fair use? I go to ChatGPT and ask it for a logo for a company with a name , it produces a copyright and trademarked image being used as logo for a similarly named company . If I go on to use that logo, is this fair use? Why is chatgpt able to reproduce this logo in this circumstance if it wont reproduce it if I ask to directly?
How is it fair use for a literal copy of this logo to exist somewhere in the dataset inside chatgpt and be produced on demand, without attribution or copyright notice, to a query, and presumably be used as a basis for producing logos in general in answer to similar queries.
>No. Learning from a book is not a derivative work covered by copyright.
Training an LLM is not "learning from a book", it has a definitive output, the resultant model updates caused by the training, which would not exist in that form without the explicit input provided by the book.
Summarizing a book "in your own words" is allowed. Do LLM have "their own words" if they exist solely as computer model that uses other peoples words as a model?
Statistical analysis of a writers style is not a derivative, having an unlicensed and illegally obtained copy of the writers work in memory to do such analysis on the fly is a problem.
>If I go to Amazon Mechanical Turk and ask for someone to tell me every 15th word of a Harry Potter book, I don't believe that would be covered by copyright either. Why should we assume that if >ChatGPT can do that it is breaching copyright?
1) Really depends on what you did with that "every 15th word" output.
2) If you went to mechanical turk and the person providing the answer went to a dark web library and downloaded a pirate copy of harry potter to relate back every 15th word? Yes, there is copyright violation happening there.
3) It implies that ChatGPT has access to the full text of this book within itself. Unlicensed and illegally obtained and presumably replicated whenever they duplicate the model
The (commercial) service relies on "training data" which is loaded into an LLM framework which stores that information is a way that isnt immediately decodable by humans, but does allow the LLM to produce large amounts of the data verbatim, at least until openAI spent quite a lot of effort over the last 12 months to handicap their own "AI" to prevent it producing said copyright works on demand.
Asked to produce a logo for a company called "netflix", chatgpt refused on the basis this would be copyrighted.
Asked to produce a logo for a company called "netflux" (no more information given at the prompt) , an extremely familiar red and black logo was produced without any refusal (at least this was true a couple of months ago).
Asked , about 6 months ago, to write a story about an orphaned boy who goes to wizard school , a few paragraphs containing very familiar elements, such as a fractional platform at kings cross station, messenger owls etc. were produced. Recently it now says something like "This sounds very similar to the premise of the 'Harry potter' series of books which are copyrighted so I cant help you".
Clearly in the hopes of heading off these inevitable lawsuits, they are attempting to obfuscate the reality of what the LLM could do.
If the premise is that the LLM, only being a computer program, does not have the "spark of creativity" that would be attributed to a human , then all of it's outputs are "by definition" derived directly from it's inputs. If those inputs are covered by copyrights, then the outputs are derivative of those copyrighted works. Apparently many of it's inputs are covered by copyright. and they have no special licensing agreements, and are hoping that just grabbing everyone else's stuff for free, storing it in a big database , and then commercially charging for an obfuscated form of access to this database is somehow covered by "fair use".
Outside of baffling the legal system with bullshit or outspending the raisers of the lawsuits, they shouldnt really have a leg to stand on.
So,you replaced a contract that would legally assign those rights to you , with a combination of a contract and an NFT , where the rights are assigned to the NFT, not you. Assuming the courts would recognized such a contract, and with the risk that someone might steal your NFT by getting your private key and legally you wouldn't have a leg to stand on? (If you could argue the NFT was stolen and therefor the rights still belonged to you then the NFT is legally worthless). Not a huge benefit, and the step where the rights are assigned to the NFT is no more secure/guaranteed than any other contract. e.g. Want to buy this NFT of the Golden Gate Bridge?
I make it about ~333 grams of hydrogen (at STP) , or 3 liters of water's worth
Obviously that isnt extracted from some mystical zero point source of moisture in the single m3 of the extractor. These things would have to be widely separated to ensure they could not interfere with each other. They would almost certainly affect the environment , making it even dryer, which means the effectiveness might deteriorate over time as well
The suggestion I have seen is the influence of Poul Anderson's 1953 novella "Three Hearts and Three Lions" which had a scottish accented dwarf, this influence went on to affect D&D and then other fantasy authors presentation of dwarves.
Much the same way as the archtypical pirate accent is a West Country one due to a single individuals portrayal of one.
You have to send a follow up command to confirm the changes within a timelimit, or they revert automatically!
So if the network is toast, then you get it back automatically after (back out testing permitting).
Plus automation setups that dont cross failure domains, ofc.
"Lobbying groups put you in a bind by hiring lobbyists who are spouses or kin to the people actually in Congress, meaning anything that threatens to blocks lobbyists will threaten critical family connections as well, creating a moral quandry."
What, the quandary between accepting dubious lobbying cash or speaking to your parents/siblings? Just like all those people torn between lucrative criminal activities or lower paying legitimate work eh?
indicates that watching *other* people play games on you tube has replaced actually playing them.
So I'm not sure why Google is worrying about pandering to the game playing minority on the same site.
Maybe the numbers of streamers are dropping, and the controller will be cunningly designed so that you cant avoid accidentally pushing the "stream my game" button?
Completely away from the terrorism aspects, if your (windows machine attached) mouse starts to do this you can implement a workaround using autohotkey until your replacement shows up.
LButton::
If (A_TimeSincePriorHotkey < 100) ;hyperclick
Return
Click Down
KeyWait, LButton
Click Up
Return
Just hope terrorists won't find this useful if their mice break, otherwise you are in trouble.
"1. The whole anti-semitism thing was satire taken out of context. Felix was making the point that people on the internet would do anything for money, even be anti-semitic. He chose anti-semitism as the most outrageous and horrific position; clearly he does not hold it himself."
He literally did it himself. To get views on his monetized channel. He is one of the "people on the internet [that] would do anything for money, even be anti-semitic".
Don't let the cognitive dissonance smack you in the ass on the way out.
And , as in the case of the comment you replied to , confusion between the product itself and the thing it manages:
"all goes well until you might actually want to speak to different tTeams"
As the OP is clearly trying to communicate with multiple clients, each with their own Teams setup.
Maybe a third party app could help:
https://jeffangama.wordpress.com/2018/03/23/how-to-connect-to-multiple-tenants-microsoft-teams-the-user-friendly-guide/
I would propose that the difficulty of "hacking" an election is proportional to the amount of work taken to run it. If your election requires printing out 10's of thousands of ballots, getting everyone in to fill them in manually, then employing 100's of people to count them, it's going to take a determined and well manned (and local) effort to interfere with that system in any significant way.
If your election set up and counting require a few mouse clicks by "Bob" the election official, then I imagine you are also only a few clicks away from massive election fraud.
I cant think of a foolproof way to even put in an "inline" audit trail into electronic voting, as even a mechanical one would rely on the selections on the screen corresponding to the correct audit output, and so could be potentially manipulated on a per-voter basis. esp. if the election required multiple options to be selected on different screens (earlier selections could be used to tell what the later selections are likely to be). Hmm, maybe you could video the whole thing and store it locally and send that in to be compared with the audit trail...of course doing that would probably take more work than using paper ballots in the first place, plus loss of vote privacy.
Once we start letting people vote with their smartphones, then Google can just decide who wins elections.
If you are going to go down that route you could probably just save a lot of time and money by asking Google how an election in any particular area would go , based on all the data they have on everyone, then use that result.
She could spend her time usefully getting the GDPR enforced against all those websites with the cookies and their ridiculous/impossible/non-existent or frankly laughable opt-out systems (like, to opt out of us collecting information, please go to some other website and opt out from there by putting a cookie on your browser for each ad-system, assuming you have cookies stored, java script enabled, and that all the ad providers systems are working at that time...), when in fact they have to be offering opt-in and opt-out is the default and should require you to do nothing.
Never mind the ones tracking everyone everywhere across the internet with their pixels etc. and no sign of any opt in or out options at all.
"What we do tend to do (and get funding for): test only the hypothesis that the observation is somehow wrong. And sometimes get into very dodgy science by proposing - say - a whole new particle to explain it."
So, erm, your complaint is that either they rigorously try to eliminate any source of error with the observation OR that they then attempt to update their theories (by "inventing new particles")?
So I assume your suggestion is that if some observation is seen that doesn't agree with the theory we should redo all the observations we have previously done that did agree just to make sure they still agree?
Pretty sure that probably isn't worthwhile until you are really, really sure the observation that didn't agree with the theory is likely accurate. And then you would be better off trying to replicate that observation (that doesn't agree), and drawing up new theories based on it that make predictions about further observations you can make/test that would disagree with the old theory but agree with your new one.
Once the theory is proven incorrect, then it's definitely going to see some action around finding out where/why it fails, but that isn't going to happen until it's really, really sure it's wrong.
Relativity is constantly being used/tested,( in GPS for instance), which shows it's reliable enough that any issues with it are going to be down to some edge case or on some scale beyond the everyday (tiny/massive distances)
"I doubt anyone at the hotel was monitoring all the guests Twitter accounts,"
Want to bet? (phrasing relevant)
I suspect they have a system with filters like "vegas, break bank, cardsharp, cheat, rig, sure thing, caesars palace" that picks out relevant tweets from all of twitter (you can do this yourself on sites like twitterfall) , flags up relevant tweets, and then checks the name/handle against current and upcoming guests . Wouldn't take much work to add "gun, attack, shoot" etc. to that existing system after the previous shooting.
Similar very recently.
Set a password (randomly generated).
Copy and paste same password into login box - doesn't work?
Read login FAQ :
Passwords cannot contain quotes(")
Then WTF did you
a) let me set one with a "
b) put "must contain a special character such as a symbol" in the listed rules , but not point out that excludes "
Not to mention it implies your back-end is vulnerable to injection, and your covering it up with sticking plasters.
In anther case, putting "#!/bin/bash" as part of a long password worked for the game login, not so good on the website as it was eventually blocked by the websites IPS as a potential injection attack... The password change tool was on the website...
From
https://www.gov.uk/government/organisations/hm-revenue-customs/contact/money-laundering
Report suspicious activity
Call HMRC if you’re an individual who needs to report suspicious activity in relation to money laundering.
Telephone:
0800 595 000
Opening times:
24 hours a day, 7 days a week
"This is going to be a fight. The telecoms and cable companies fight hard and they are effective. We have our work cut out for us."
Well, lacking any law to the contrary , the telcos and cable companies can block access to pro-net neutrality websites, redirect customers to websites promoting negative stories about network neutrality supporting politicians, replace all your adverts with adverts for how much better things would be without net-neutrality.
Also bin any emails to your representatives that support net-neutrality of course.
"I wonder if anyone hasn't "VW-ed" their array"
FTFY
The SPC benchmarks are massively gamed and always have been.
Commercially nonsensical hardware optimized specifically to the benchmark.
Loads of controllers with tiny amounts of disk each to maximise the cache memory available.
Volumes made from just the fastest part of traditional disks , while still quoting the full capacity in the $/GB.
Specialized firmware setups are just the tip of the iceberg.
"gig that later went very pear-shaped indeed (although the client was to blame)"
I wonder how IBM got that previous contract?
"“The only finding possible is that IBM should not have been appointed” to the contract in the first place in part because of “ethical transgressions” on the part of some of its employees, including “the obligation not to use the State’s confidential [bid and proposal] information” that it had somehow couldn't explain came into its possession from a restricted government database along with the apparent privileged insider information from a government consultant to the project who happened to be a former-IBM employee."
Any former IBM employees involved with this new bid?
Ancient wise man 1: so, that's counting sorted out then
one, two three, death, five, six ...
Ancient wise man 2: I'm sorry, three, what?
AW1: Death.
AW2: Why is it one two three, then death?
AW1: It's just what we came up with. Of course, probably people will try to avoid having "Death" amounts of things I suppose, maybe a little creepy, but what can you do eh? Anyway.,
Seven, eight, nine, ten, eleven, twelve, thirteen, deathteen, fifteen.
AW2: Erm, deathteen?
AW1: Well, we have to be consistent, otherwise people would get confused. We also have twenty-death, thirty-death and my favourite, deathty-death.
.....
Try to imagine it in a Mitchell and web style.
"did she have the training and education to be doing this?"
There are people who have roughly the training and skill set to do this job, they are called "Driving instructors". Not hiring someone with equivalent skills means that Uber doesn't give a crap about the effectiveness of the person in the seat, it's just a cost they have to pay to meet the minimum legal requirement.
Their immediate response, drop all blame on driver, and run away to another state.
The problem is that although they should be equally liable (for bad software, lack of supervision, insufficiently skilled employees), they will likely get away with it as the legal status of experimental self driving cars isn't being given sufficient attention.
"address allocation optimization requirements for IPv4 bear no relation to sensible and relevant optimization strategies for IPv6."
https://tools.ietf.org/html/rfc6177#section-2
Hence, this document still recommends giving home sites significantly more than a single /64 , but does not recommend that every home site be given a /48 either.
/64 1 IPv6 subnet 18,446,744,073,709,551,616 IPv6 addresses
Now, I'm not sure how much IoT shit I'm supposed to put in my house, but 4 billion ipv4 internets worth seems like a lot? Maybe they could have increased the lifespan of ipv6 by only giving homes 1 ipv4 internets worth.
If the minimum subnet size is 2^64 , and the complaint is "the routers will fill up if we have millions of routes". how exactly are switches going to cope if you put millions quintillions of hosts on one subnet? It's all just wasted address space at the cost of much longer addresses.
Of course, ipv6 may give you quintillions of routable ip addresses but only has one loopback address.