The legal fight between OpenAI and The New York Times over data to train its AI models may still be going on. Of course, OpenAI is moving forward with deals with other companies, such as some of the biggest news outlets in France and Spain.
OpenAI said on Wednesday that it had signed deals with Le Monde and Prisa Media to add news from France and Spain to its ChatGPT robot. OpenAI said in a blog post that the agreement will bring current events coverage from brands like El PaĆs, Cinco DĆas, As, and El Huffpost to ChatGPT users where it makes sense. It will also add to OpenAI’s growing collection of training data.
OpenAI Says This:
In the coming months, ChatGPT users will be able to connect with relevant news content from these publishers through select summaries that include attribution and better links to the original articles. This will allow users to get more information or related articles from their news sites. By making improvements to ChatGPT all the time, we’re helping the news industry’s important job of giving users accurate, up-to-date information.
The company OpenAI has now made distribution deals with a number of content providers. It seemed like a good time to take stock:
- Library of stock media Shutterstock (for pictures, movies, and music video training)
- The News Service of America
- It is owned by Axel Springer, who also owns Politico and Business Insider.
- The Le Monde
- Press Prisa
How much does OpenAI pay each person? It’s not saying, at least not in public. We can guess, though.
In January, The Information said that OpenAI was giving publishers access to its records for between $1 million and $5 million a year so that it could train its GenAI models. That doesn’t tell us much about the deal with Shutterstock. However, OpenAI pays between $4 million and $20 million a year for news articles, believing that what The Information said is true and that the numbers haven’t changed since then.
That might be very little money to OpenAI, which has a war chest of over $11 billion and recently made more than $2 billion a year (Financial Times). But Hunter Walk, a partner at Homebrew and co-founder of Screendoor, recently thought that it might be big enough to beat out AI competitors who are also looking for licensing deals.
On His Blog, Walk Writes:
[I]f nine-figure licensing deals stop people from trying new things, we’re not helping innovation… The lack of checks on “owners” of training data is making it very hard for rivals to get in. If Google, OpenAI, and other big tech companies can set a price that is high enough, they will stop other companies from competing.
Now, it’s not clear if there’s a barrier to entry today. Many AI vendors, if not most of them, have decided to risk the anger of IP owners by not licensing the data they use to train their AI models. There’s proof that the art-making tool Midjourney is training on stills from Disney movies, even though Midjourney doesn’t work with Disney.
Maybe the tougher question is: Should licensing just be the price of doing business and experimenting in the AI space?
Walk would say no. He wants the government to set up a “safe harbor” that would protect all AI vendors, including small businesses and researchers, from being sued as long as they follow certain rules of honesty and integrity.
A strange thing happened lately in the U.K.: they tried to make a law that would allow text and data mining for AI training without worrying about copyright issues as long as it was for research. But those plans didn’t work out in the end.
I’m not sure I’d go as far as Walk does with his “safe harbor” plan, since AI could make an already unstable news business even worse. The Atlantic recently did a study that showed if a search engine like Google added AI to its search, it would answer a user’s question 75% of the time without them having to click through to its website.
But Maybe There’s Room For Exceptions.
It is fair for publishers to be paid what they’re worth. But wouldn’t it be great if they got paid and AI challengers and researchers could both access the same data as the AI challengers? I believe so. One way is to get grants. One more is bigger VC checks.
Also Read: Openai Does Not Have a āgptā Brand
I’m not sure what the answer is because the courts have not yet decided if and how much fair use protects AI providers from copyright claims. But it’s important that we figure these things out. Because if they don’t, the industry might end up with a lot of “brain drain” from universities and only a few big companies having access to a lot of useful training materials.
What do you say about this story? VisitĀ Parhlo WorldĀ For more.
