It appears that there are at least two legal questions: (1) Was ChatGPT trained on improperly obtained copyrighted material and, of so, how does that taint any generated text? (2) The one you incompletely state: Even if it was trained on legally obtained material, is the generated text (a) derivative, (b) transformative, or (c) neither transformative not derivative but simply novel work. (2)(c) is important. If I read a bunch of J.R.R. Martin and then write a fantasy story where people say "I was afeared" and there are dragons, knights, kings, and magical beings, but no Westeros or Tyrion or a wall of ice, it would be (2)(c) even though I learned from Game of Thrones. As an AI in cognitive neuroscience expert, but not a legal one, I have to ask a simple, yet profound question implied by Lukyo above: What is the difference between text generated by AI and text generated by a human who has read lots of novels, web pages, textbooks, etc.? I have seen no compelling argument that it is different. All the arguments that purport to do so ultimately evoke some magical fairy dust argument that "humans as special" with no evidence. Simply evoking non-text-based experience is insufficient, since that argument, even if true, would be a dilution argument: If "only a portion of my work was influenced by copyrighted material" is a valid argument, then what percentage of the training can be unauthorized and if we establish a percentage, then how do you determine what percentage was used for any given text generated by an LLM (spoiler: you cannot)? I am an author of textbooks, scientific papers (those most of those I had to give up my copyright), and a number of articles in magazines, newspapers, and the like. I understand the concerns, but I cannot find a valid argument that does not either evoke magical thinking for the special nature of humans or which falls apart logically. That does not mean, of course, that we won't see a legal ruling that does not map onto science and logic.

Damn, I stubbed my tow while typing this. I should use ChatGPT next time.

Folks: rpnx linked to what he implies is clear evidence. Read it before you upvote, for crying out loud.

It looks to me that the evidence cited above is anything but direct evidence. I'm sure this is not the only evidence or documents turned over and/or submitted, but so far, no "smoking gun"..

This does NOT mean that the government did not act inappropriately. It may have, and perhaps likely did; however, rpnx seems to have either linked without reading or is being disingenuous, or maybe the smoking gun is in the 7/20 I didn't read through. See the summaries below for some of the items. I'm not doing them all, but the theme persists..

In short, it's clear that some people's social media accounts were suspended or otherwise censored. It's also clear that some government scientists were unhappy with some of the information and asked for a decisive rebuttal (which is NOT asking for censorship) from within their ranks. There is no clear evidence I can see so far of a governmental agent, agency, or scientist, directly pressuring any person or corporation to censor any one individual, corporation, group, or publication. I have not read all of the linked PDFs in detail, because I have a life some days, but the 13 of the 20 I did read thoroughly seemed devoid of any "smoking gun" evidence of government interference. For example, in the last doc, there is evidence that Facebook sought guidance from the government on what constitutes misinformation and that the Gvt was eager to comply. I am unaware of any caselaw that would flag this as governmental censorship. If I, as a private citizen or corporation, want to toe the line of sitting members of my party or government officials I like, however sound or crazy, and seek their input on what to promote or quash, I am 100% within my right to do so and they are within their right to tell me what would make them happy or less so. That would be my 1st amendment right..

We do know, at least that's what we've been told since 2020, is that the government has been actively trying to combat misinformation. Combatting misinformation, particularly misinformation that can cause harm, is important for a government to do; however, who decides and why is always problematic. There is quite a bit of caselaw, however, stating that the government may, in some limited circumstances, limit free speech, so we'll have to wait for this whole thing to play out to know more of the details..

Summary:.

[1] Some random email from someone at FaceBook (Meta) asking for CDC copy edits. 100% OK for FB to do. No evidence of any Gvt. coercion..

[2] Affiant claiming to have been targeted by Gct. coercion, but offering no evidence..

[3] Prominent scientists claiming active Gvt. coercion, but offering no evidence. "Smoking gun" emails show government scientists (Francis Collins) saying they need "a quick and devastating takedown" which, by the affiant's own admission, never happened and even if it did, it's in the government's rights to rebut even in a "devastating" manner. More interestingly, Dr Bhattacharya makes claims about "numerous studies" (which he should know very well as they would be in his field), yet fails to provide a single reference in the 11-page affidavit. Both attest to having been censored, but neither have direct evidence that it was the government..

[4] The same exact thing for Dr. Kulldorff's affidavit. They are likely cut-and-pasted from one another. and so on..

Note: Regarding Dr. Bhattacharya and Dr. Kulldorff's affidavits: the one time I filled an expert witness affidavit, I had 103 references to peer-reviewed studies, many not in my field. I find it very odd that these notable scientists are waiving around blank papers saying "many studies". Which studies? It's your field. You should personally know some of these authors if you're not one of them..

Setting one pixel to black is not 100% fool proof unless you have only one pixel. There is a whole subfield of research, some of which uses AI, that attempts to reconstruct images that have been blacked out. There are some that do a surprising job under many circumstances; however, they are never anywhere close to 100% accurate and if you block out a whole face, for example, some of the AI systems will correctly put a face in, but it's a generic face which would be essentially useless for tracking some perpetrator. Although CSI "Enhance" is pure crap, some AI enhancing is pretty remarkable.

No because (1) very, very few technologies require export licenses, and (2) when they do, it's not the DoD or the DoJ that issue licenses. If you are referring to the inability to export strong cryptography, then with very few exceptions (and, yes, made through the DoD to enable disseminating to close allies for military/government use only) you can't, whether you're Apple or John Doe. There is no "license" to allow you to export 40,000-bit, quantum-computing-unkrackable encryption -- it's just illegal. This has been circumvented by having (1) your dev team outside the US, so you are importing it into the US (which is currently largely legal, contrary to what you imply above), or (2) making the bit-depth configurable as PGP does, so what is exported is "legal" but when used it's configured to something stronger than would have been allowed.

Meh. Only an issue if the government is your main or only target audience, in which case, do you individually really care if you bought something intended for the military and you find out that they can spy on you? (Note: having a backdoor on military tech is an even dumber idea, but we're not arguing if the government is always doing the smartest things). Apple, Google, Facebook, and even Microsoft are happy to play with the G-men, but none would abandon their 90% of non-government market

National security letters are the devil incarnate and can kill a startup and hamper almost any size company, but the FCC has nothing to do with them. Ajit Pai is a festering pustule poised to explode on the US's behind for a lot of good reasons, but not this one.

If by "real work" you mean work useful outside of the blockchain use case, the short answer: "not that we know of."

In order for the blockchain cryptocurrency ecosystem to work in a decentralized, zero-trust environment as intended, there cannot be any direction on what constitutes proof-of-work by a controlling entity (e.g., a person or organization). In order to create value in the currency and disable the ability for anyone to "print new bitcoins for free", new currency is created subsequent only to proof-of-work. (there are great videos that will explain this better than a short /. post).

How do you validate proof-of-work if there is no final arbiter or central bank? What is needed is a reliable, (somewhat) arbitrarily defined, increasing complexity task which poses a computational question with an answer which can be proved correct (or not) with few computational resources. Although we don't (and must not be able to know before hand) the "answer" to the question before hand, the class of computational problems of which we know that fit this paradigm are found in the cryptography or chaotic function realms (e.g., hashing functions).

Our current knowledge suggests that we probably wont find a class of problems that (1) is really hard to answer, (2) is really easy to check, and (3) the answers themselves are useful for solving real-world questions.

Actually, you missed the point entirely of the first question and response, since no one was asking "hey, should I start mining Bitcoins in order to heat my house?"

IF you are generating heat through computation, using it to heat your house is a significantly better solution than not. Period. Not using it, but then using a "more efficient" heat source makes things considerably worse.

The heat generated by computing is waste heat. If it is put to good use, it's a good thing. So, if you're going to create a lot of wast heat, you're doing the world a favor when you make good use of it. The problem is that "cryptocurrency" requires computationally intensive but otherwise (outside of the currency market) useless and meaningless work as proof-of-work. If that work were also socially beneficial (e.g., curing death), then using the excess heat to warm your LEED certifies Alaskan cabin would be a great thing all around.

You are successful in getting high scores on slashdot saying essentially the same thing (https://hardware.slashdot.org/comments.pl?sid=10552741&cid=54327851) so you're likely getting some part of your point across to some readers. That being said, saying that "we're nowhere close to AI" is misleading at best. "We are nowhere close to AI that resembles or acts like a human" is absolutely true; however, the AI that is learning about you right now is NOT an "expert system". Not even close. Expert systems are completely constrained (unless augmented with neural networks) by human programmers. The "sexy" AI we hear about on a daily basis are very much a black box, often with aggregated of artificial neural networks (ANNs). The fact that ANNs can be largely reduced to a set of discrete numerical outputs does not limit the complexity of their abilities at all. Bit have two discrete outputs but can represent unfathomably complex numbers and instruction sets when put together.

Blind the code reviews. Done.

If it's a matter of lower quality, address that, whether it's because you constantly have junior females (they're leaving or you're just ramping up, for example) or not, you remove any bias in the code review. It's not rocket surgery. If you can't blind them all (doubtful in a company the size of FB) then blind some randomly. If the difference goes away (as does happen, see Google studies and Harvard studies on code reviews and acceptance rates when gender is blinded), the problem solved, at least for code. Move on to next problem.

This.

I don't understand why more customers and retailers are not screaming bloody murder over this. Retailers are losing billions because they either have to staff more checkouts and have longer lines (and irritated customers). Customers should be pissed off since they wait longer, have to do more (insert, wait, pin or sign or do the hokey pokey, remove the card when you get that fart sound).

The United States is now the country of security theater that pisses everyone off (e.g., TSA) does nothing to increase security (read the DoD and Homland Sec's own reports) and everyone seems to accept it. God I love and hate my country....

Elsevier just says that original work created by the US Government does not need to transfer copyright to Elsevier for that original work in order for it to be published by Elsevier in one of the their journals.

What you cite is completely irrelevant to what people are worried about.

Elsevier does not allow anyone in the license to re-publish work of others which is apparently what this law would require in order for the EPA to consider any science.... therefore .... wait for it... the EPA would still not be able to use 99.99% of published science.

What Internet were you using? USENET, dial-in BBSs, and even MUDs were full of self-righteous condescending jackasses, trolls, and ragers well before AOL. I missed the unicorns and fairy Internet train I guess...

Unfortunately, there are no current or proposed market models that are shown to work (in simulation or in the real world) when a significant portion of the population is no longer among the producers of goods or services.

This is a huge problem. Remember that for the entire existence of any economy, whether a barter economy of 2,000 BCE or today, a majority of the population needed to be among producers of goods and services. Without that, the markets collapsed.

Another foreboding historical lesson is that when unemployment crosses 20%, social unrest increases, murder rates go up drastically, depression and suicide spike, and happiness tanks. PwC, Forbes, Gartner, and many others forecast between 35% to 47% jobs at risk over the next 7-15 years. That's around the corner, so let that sink in. And a recent poll found that the vast majority of workers think automation will take "someone else's" job. If you think it's "not my job" when AI code writers and debuggers are being deployed, white collar workers are in the cross hairs, and all manufacturing positions are at high right, you might want to pull your head out of the sand.

But supposedly we're going to barrel past 20% unemployment. So what has history taught us about unemployment passing 40%? Well, every instance has resulted in violent social unrest with at times up to 25% of the population being murdered. That does not sound too rosy.

OK, so universal basic income it is, right? Not so fast. See above. No economic model works even with UBI. What happens to inflation when 45% of adults don't do anything for their money? See, there is no good solution on the table.

The only thing pretty likely is that if you're worth well over $10M to $15M USD then you and your immediate family (of spouse and two kids) are probably OK for the next quarter century. That's less than one in 10,000 of the US population and less than on in about 10,000,000 of the world population. That's a lot of pissed off destitute people that might start popping up everywhere.

So if it gets bad, we'll have a revolution! *sigh* Good luck revolting against an AI augmented, robot-assisted, militarized police force (or private security) that knows who you've been talking to about what, where you are at any given point in time, and what you have been learning. Remember, the House and Senate just allowed more of your private data to be scooped up and monetized, and the US Constitution used to protect you against the Government invading your privacy, but never protected you from private citizens or companies (or private security). Good times ahead for all if you don't get involved NOW. 2020 is too late and AI is accelerating all timelines.

P.S.,(NOTE: Tinfoil hat time) if you think that you can just leave your burner cell phone at home, remove all the RFID tags in your clothes, credit cards, and grocery store discount cards from your person, and cover your face, alter your gait, change your height (all things used for automated recognition) or otherwise live off the grid, let me burst your bubble. Research from Stanford, MIT, WUSTL and other places show how many ways the people around you with all that stuff still active can make you trackable even if your were only wearing a gorrilla outfit. In short, don't get depressed, get involved. NOW. Trump, the alt-right, the Dems, the Neo-cons, the Commies, and even the Libertarians are not going to fix this.

Although there is research showing an increased prevalence of some general social dysfunction disorders (e.g., NPD and others which are often colloquially referred to as "sociopathy") among certain classes of occupations, I can find no serious research on "psychopathy" (not a real disorder or class of disorders) or any other empathy-deficient disorder with respect to their disproportionate prevalence in executive leadership. It intuitively seems like it might be likely, but anyone can pull that out of their ear. This is supposedly the expressed opinion of one or more panelists, none of which are even quoted directly and only two of which might claim any expertise in the area (but neither of which have published on this particular matter). In other words, sensationalist click bait.

P.S., I have opinions on this too, and they are worth about as much.

If this is the same suit filed against many other academic institutions, then the plaintiffs allege that the automatic captioning (among other things) is inadequate today. I will take that argument as true, though I would dispute it at least in part. Regardless, It is not relevant whether or not it will be one day adequate, as the ADA does not say it's OK to have an inaccessible building because "we can fix it later". The person needs to access the building / bathroom / course material / you name it now. This makes perfect sense until it has serious negative consequences for everyone.

The problem with the ADA is that it does not work as intended in this particular case. Under no circumstances would anyone say, "it costs too much to make all buildings ADA compliant, therefore we should just abandon all buildings and no one will have them." But when you have a tenant with special needs, it is reasonable to require that accommodations be made. Unfortunately, there is no reasonable equivalent to large-scale "giving away stuff for free" and so the ADA puts people in a horrible predicament if taken literally -- the predicament in which UC Berkeley among others finds itself.

The ADA expectation is that the content be accessible at the time of need and that it be made so by the institution. I, unfortunately, worry that this kind of suit will have to tremendously negative effects: (1) it will penalize the vast majority of the public because an important but still tiny minority cannot be granted the same level of access and, perhaps more importantly, (2) will seriously damage the public sentiment toward those less privileged, who quite seriously, do not need this. The latter should be of high concern to all ADA advocates. Basically, it will infuriate millions and further alienate other millions who are disadvantaged. Lose-lose if you ask me.

If the objective is to ensure wider access to those who have special needs, this not only does nothing to further that goal but damages future prospects of achieving that goal for everyone. Most of these institutions (especially the non-profit institutions) want to make the material available to everyone, disabled included, and I refuse to believe there is not a community-wide potential solution that is not universally damaging.