OpenAI, Axel Springer Strike Unprecedented Deal To Offer News In ChatGPT (cnbc.com) 23
OpenAI has struck a deal with Politico parent company Axel Springer, allowing ChatGPT to summarize news stories from Politico and Business Insider. CNBC reports: Once the OpenAI-Axel Springer deal goes into effect, when a user asks ChatGPT a question, it will respond with summaries of news articles from media outlets such as Politico, Business Insider, Bild and Welt. The chatbot will also include articles that would otherwise be limited to subscribers of those outlets, according to a release, and the answers will include "attribution and links to the full articles for transparency." The partnership follows a deal that OpenAI struck with the Associated Press in July, allowing it to license the AP's news archive for training data.
As part of the agreement, Axel Springer will provide content from its media brands as training data for OpenAI's large language models, such as GPT-4, the AI model that helps power ChatGPT. The News Media Alliance, a trade group representing more than 2,200 publishers, released research in October suggesting that data sets used to train popular AI models rely "significantly" more on publisher content, outweighing it by a factor ranging from over five to almost 100, compared to generic web content.
As part of the agreement, Axel Springer will provide content from its media brands as training data for OpenAI's large language models, such as GPT-4, the AI model that helps power ChatGPT. The News Media Alliance, a trade group representing more than 2,200 publishers, released research in October suggesting that data sets used to train popular AI models rely "significantly" more on publisher content, outweighing it by a factor ranging from over five to almost 100, compared to generic web content.
will ChatGPT make them up? (Score:2)
When they were training up ChatGPT, did they teach to tell what is fiction? imho the fake cases cited were from tv/movie scripts they fed it.
Re: (Score:2)
Trump willbe the first AI ChatPresident
Nope. Ramaswamy is the current AI Chatbot running for office. Trump always sounds disconnected, unhinged, maybe in need of a handler. Ramaswamy can go from semi-coherent, almost literate sounding talking points that are posted everywhere on the internet, to completely unhinged lunacy that would make grandpappy construction worker blush, that are posted everywhere on the internet. And sometimes one sentence to another makes it seem like he doesn't even really connect the ideas he's spouting, much like a chat
Re: (Score:2)
Like it has been caught making up court cases when asked for legal briefs?
That happened almost a year ago. LLMs have improved a lot since then.
When they were training up ChatGPT, did they teach to tell what is fiction?
Newer LLMs are getting better at that. If you ask about well-known fictional stories, they can tell you they are fiction.
Re: (Score:2)
LLMs have improved a lot since then. [...] Newer LLMs are getting better at that. If you ask about well-known fictional stories, they can tell you they are fiction.
Not really. How do I explain this? To say that there have been improvements made that better allow modern models to identify fiction is to imply that there exists some analytic process by which the model could identify input as fictional and that improvements have been made to that process. That's impossible. These things don't operate on facts and concepts, they operate on statistical relationships between tokens. That's all. They do no and can not operate analytically.
The "improvements" you're seein
They don't exactly publish news (Score:2)
Axel Springer is not exactly a news company, their business is more like advertisements disguised as news. For example they constantly write against heat-pumps without providing any facts. They also had front-page articles agitating against every measure to help condemn the spread of Covid19... going as far as to personally attack the minister of health.
This is a terrible idea (Score:4, Insightful)
These LLMs already have a problem with authoritative bullsh*tting. We don't need even more of that being injected into the newscycle.
Re: This is a terrible idea (Score:2)
A summary of a summary. (Score:2)
Re: (Score:2)
We'll continue until we can summarize news into a single emoji.
Re: (Score:2)
I wonder if talking heads (Score:2)
I don't know (Score:2)
As someone who works in media and is a member of a union affiliated with the News Guild of New York, this sounds shitty. But I suppose that if anyone wants to use a bot to summarize news stories, I guess the bot needs to have something to summarize, which presumably would be written by people.
Misread as Jerry Springer (Score:3)
ChatGPT... You are NOT the father.
What about the lying? (Score:3)
Re: (Score:2)
ChatGPT is infamous for lying assuredly and convincingly.
Coincidently, so is Axel Springer Verlag. Truly a match made in hell.
Bad input for ChatGPT (Score:3)
Bild is the biggest and worst tabloid in Germany: cheap, dumbed down language, big headlines, sensationalism, sex sells, almost no facts checking.
Axel Springer in general is known for biased content.
ChatGPT reading the news that it wrote (Score:3)
based on information that was probably also written by it.
Even a minor bias in the algorithms will ChatGPT for full Tay on it's users after a few days.
AlexSpringer and DPG Media are media cesspitts (Score:4)
Their 'hidden' agenda is so obvious that these are not media outlets but propaganda outlets.
What Could Go Wrong? (Score:2, Informative)