Grim. This is just confirmation that the future of llms will be a walled garden of big tech and big rights holders. There's no open ecosystem if content needs to be licensed at prohibitive prices for small players/ universities/ open source. Between this and them lobbying for regulation closed ai is obviously pulling up the ladder behind them now that they're valued at $80 billion from being reportedly on the brink of bankruptcy just a few years ago before chatgpt

What's the alternative?

Do you feel that entities like The Financial Times should just willingly or be forced to give up all their data to the public?

That was the intent of copyright, until it was extended from 14 years to beyond the heat death of the universe.

Maybe we should go back to 14 years for use in AI models?

Sure. Just like it always was, like when libraries buy magazines and papers and archive them, giving the public access to their contents.

The difference is that mega-corps are ripping off content and reselling it. I hope that the FT can sustain itself for a long time using the money from this deal.

OK, so you think this is something new? There are no examples of mega-corporations "ripping off" and "reselling" from, say, the sixties?

Yes.

You should ask yourself how you reached the point that you're fighting on big corp side? I find that extremely interesting.

Most propaganda is obvious. But this... This is amazing. No media clearly puts it on a pedestal. But, so many tech people I talk to have this knee jerk reaction to vehemently defend whatever large corp is in the context. Amazing.

Anyway, back on point; Yes. Obviously. It is in my best interest that information be as accessible as possible. So, I'd like as much of it as possible to be as accessible as possible. There is a law here that forces marketplaces to publish their prices of all their inventory at any given time in an accessible digital manner. This is a push in the right direction. Must it all be black and white; Does TFT have to work for free? No. But whatever solution there is to keep data accessible is in my best interest. Their revenue or their industrial advantage is not in my best interest.

Do you work for free? If so - if you are a volunteer and do not take state aid or use funds from your inheritance - then your position is defensible.

If not, then I invite you to look at the huge attrition of journalism jobs in the UK and the USA. Local newsrooms have been destroyed by the web and national newsrooms have been gutted.

Folks that create content should be allowed to monetize content - in exactly the same way that I am allowed to monetize the work I do.

They are currently allowed to do so, and this attrition has happened without a proliferation of LLMs.

This feels like an oranges-apples kinda argument. The thing that's destroying journalism is the current framework for monetizing and how utterly inadequate it is - that, and the lack of a robust middle class; the primary consumer of news.

It's easy to point at tech and say that it's what changed so it's the problem - but just like dirt roads were better for horses, sometimes it's the existing structures that need to change. Our ownership model is lost in the modern era and badly needs revisiting.

Sounds good to me. I don't see why anyone else should feel they have the rights to content just to train crappy chat bots. Real people with real families and real mortgages created that content.

In this case is it better that they get money and rich people consolidate more power over an emerging and likely influential technology, or is it better that they don’t and open source is able to flourish?

They're only licensing what they have to under threat of lawsuit. Little people's data will still be taken without refund.

OpenAI and other big VC funded LLM companies were all destined for enshittification anyway. They'll try and regulatory capture their way to monopoly and then offer a much worse product for more money. Luckily (hopefully) the cat is out of the bag and there are enough good, geographically diverse alternatives to prevent any serious regulatory capture efforts. Worst case we'll all be fleeing to the freedom of the Chinese LLMs.

This is fine because LLMs are a dead end.

If LLMs are the core of 'AI' then I'm not interested, but I feel confident they're a sideshow on the way to stronger AI.

Lets avoid the huge mistake that was made with the internet, where the web happened to be the first broadly useful tool which exploded in popularity and then de facto became the internet, with everything shoehorned into it. The problem is, on the scale of possible interfaces and software, the web is absolute shite, but now its gravity is too great and we can't escape it.

I've tried to use LLMs but they're just not useful. Getting answers that may be great or may be nonsense just doesn't work for me. I know there is something better and I won't be distracted by a very impressive novelty.

I don't think they're a dead end, they're just a very small piece of a much bigger puzzle. LLMs are to AI what word2vec was to machine learning - it's the key to encoding language in a way that algorithms can operate on but they need to be connected to a bunch of other systems to really be useful.

I think it'll be a while before we perfect memory, neuroplasticity, the physical experimentation feedback loop, etc. but LLMs at least give us a way to represent and manipulate human language.

> just a very small piece of a much bigger puzzle

I agree with that, but plenty of people are talking about LLMs as the precursor to AGI, especially the OpenAI crowd. Plenty of people here and elsewhere say that LLMs are actually intelligent.

I think LLMs are an important illustration of what's possible, but it's not the one true way.

The publishing industry is in big trouble; journalists are fired, and publications are closing.

Ads are not enough, and readers are fed up with ads, so they use adblockers, cutting publications' revenues. With AI chatbots, people browse these publications even less, further reducing revenues.

Paywall and licensing content is the next best option, if not what it is?

Just wanted to chime in and say that I think you are both right. I haven't thought of a way to square this circle yet. I find this issue to be the fundamental concern around generative AI. Of course the Courts could always end up saying its fair game and at that point, I could see OpenAI no longer paying. And then content producers are more likely to gate content in a myriad of ways and the cat and mouse game will continue. I could also see the US say scraping is allowed and EU saying otherwise.

publishing indstury maybe be in trouble, but publishing itself is not.

content creators are going direct to consumer with their content, and there is endless opportunity for actual content creators to thrive without publishing houses as middlemen

It seems to me that half the closures happen after some wealthy ass hole that seems to hate journalists buys a publication then runs it into the ground.

If you can't run your business without violating the property rights of millions of people, maybe your business shouldn't exist.

This comment is direct, but not unreasonable. To steel man it...

The success of the LLM were built by the work product of others, without compensation. Now those others are looking for their due compensation.

Freely profiting off the, quite literally, compressed output of others isn't really a business model that's sustainable, for either side. The only sustainable solution would necessarily involves money going from the content users (multi B $ LLM companies) to the content producers (artists, news orgs, etc). For a logical litmus test, apply what's happening to any other content/industry.

Hey thats fine if you have that opinion.

I just hope you are consistent in that you oppose the existing ubiquitous copyright violations done by the entire art industry, in the form of commercial "fan art", or similar.

A whole lot of content is built off of other people's works, and much of it is not done "in fair use", but of course, then the shoe is on the other foot, those same creators complain.

I have a lot of trouble taking this obviously bad faith argument seriously. Aside from the fact that we're comparing random fan artists with well resourced tech corporations that trained on hundreds of millions of images, most rightsholders seem to appreciate that fan art is free publicity for them. If nobody is going after fan art, then there's not really a problem. On the other hand, many people and organizations seem pretty happy to file suits against OpenAI.

I'm not sure AI businesses would be viable if they had to license all the content in addition to paying all the compute costs. But, that really isn't the rightsholders' problem.

That's the clearest indication that OpenAI should be seeking permissions and paying rightsholders for training data.

If OpenAI loses value as a company if it does not have that content, that indicates the content has value, and OpenAI should pay for the materials they 'use' to create their own value.

These models will exist one way or another. Even the most IP concerned lawmakers in the deepest pockets of the RIAA and whatnot. It's just that some want there to be monopoly rents payed to themselves every time someone uses them

> These models will exist one way or another.

This line of reasoning seems more or less equivalent to "movie piracy will exist one way or another, therefore I should be allowed to launch a commercial Netflix competitor which streams movies without paying the studios". Just because it's possible to appropriate content doesn't mean it should effectively be put in the public domain, that's obviously not sustainable.

Actually that sounds great. Commercial piracy netflix competitor please and thank you.

Sounds great until the entertainment industry collapses because the inconvenience barrier which kept piracy at reasonable levels vanishes, with every theatre, TV channel and streaming service suddenly no longer paying for anything they show. I think you know that's not going to work.

Yeah, but wouldn’t it be a heck of a lot of fun

Maybe profit caps for models that use licensed/copyright content would make sense? That would leave general use of open source models alone, while still having the potential to flow money to the right people.

I can’t understand why publications would strike deals for their content to be used for training LLMs. It seems incredibly short-sighted to me. They gain a windfall in exchange for what is basically nailing their own coffin shut.

I don't know man? Ad revenue is plummeting. No one else is buying their content.

What is the alternative path that is as lucrative as taking that big 4$$ check from OpenAI/Microsoft? I mean the only real alternative is that Google, Facebook, or Amazon write you a bigger check. But you're still at the same place because they'd only write that check to train their models as well.

The industry is backed into a corner. If no one else will pay, they have to take money from the only person who will.

> What is the alternative path that is as lucrative as taking that big 4$$ check from OpenAI/Microsoft?

I can imagine LLMs tipping the balance back in favor of something like traditional media. In short: people go nuts with them, and flood the open web with garbage (which may have the added benefit of degrading LLMs). That could, compared to the last 20 years, make curation and verified provenance far more important for people who value having the chance to know real things, which would drive customers back to pay-walled media organizations and publishers.

That may not happen, or may not happen on a time-frame compatible with most current media organizations. It would certainly take some time for everyday users to develop the necessary fatigue with LLMs and their output to motivate action.

I think it'd also depend on the LLM companies losing in court on copyright grounds. If the LLM companies win, I think we'll still get the crapflood, but the islands of sanity resisting the tide will be smaller or nonexistent. If LLMs put the final nail in the coffin of the media orgs, Wikipedia will die soon after (it's got separate culture problems that may do it in, but it is also totally dependent on the editorial decisions of traditional media and publishing).

Well the FT charges £39 per month for their content, without even paying extra for premium. And they have ads on top!

If those people don't sell their content, other publishers will and they will lose out anyway.

Each individual publisher isn't needed. There is lots of training data in the world. So you can either get a payday or get nothing.

Related official post:

We’re bringing the Financial Times’ world-class journalism to ChatGPT

https://openai.com/blog/content-partnership-with-financial-t...

(https://news.ycombinator.com/item?id=40197303)

FT release:

https://aboutus.ft.com/press_release/openai

(https://news.ycombinator.com/item?id=40199571)

So does this set a precedent that OpenAI knows it should be paying the millions of rights holders who didn't grant it a license and yet saw it use their IP to generate billions of dollars of revenue?

One of the factors in fair use analysis is whether or not there exists a market for licensing for the use in question, and this is conventionally given the strongest weight in fair use analysis.

So yes, making agreements to license content does illustrate that there exists a market for using text in AI training, and it will do a lot of damage to arguments that it's all fair use.

This is not the first publisher OpenAI has paid for rights to its content.

https://www.bloomberg.com/news/articles/2023-12-13/openai-ax...

Is this just inevitable now?

I also can't believe that these media companies haven't learned the lesson they shouldn't have learned by looking what happened with Google and saying maybe we shouldn't give that away, at any cost. Like what's their bargaining position after the training has been done?

They did learn their lesson with google.

That's why they're taking the money that OpenAI/Microsoft is offering.

They tried to do it in court with google and never got anywhere with it. So this time around they just take the money from whoever is willing to write the biggest check.

that lesson isn't worth much to an industry that has constant solvency issues

I wonder if this is some long play by OpenAI:

1. OpenAI is trying (and failing?) to add legal restrictions on LLM (and co) training to only a few major players.

2. By entering into licensing agreements, it starts to set the standard and essentially forces the above to take place as only a few major players would be financially able to afford it.

yes, the concerns around limiting access to the newcomers are valid, but there's the other side to the deal:

> In addition, the FT became a customer of ChatGPT Enterprise earlier this year, purchasing access for all FT employees to ensure its teams are well-versed in the technology and can benefit from the creativity and productivity gains made possible by OpenAI’s tools.

i.e. ft is going to be written by llms now.

Just skip the middleman and have chatgpt generate financial times articles to train on

FT press release (not paywalled): https://aboutus.ft.com/press_release/openai

From behind the paywall -- NYT sued them, and so OAI will pursue for now licensing arrangements with some major media outlets at least.

This should have happened years ago, and not just with one big publisher.