openai OpenAI partners with controversial Axel Springer SE to use and train with their news content

[DerGottesgleiche]

Summary

German media publishing giant Axel Springer SE and OpenAI announced a partnership to quote: "deepen beneficial use of AI in journalism"

There are plans to have ChatGPT link and reference News Articles of Axel Springer properties like the German Welt-Newspaper, the tabloid Bild Newspaper, but also major international news publications like Politico, Upday (cooperation with Samsung) and Business Insider. The agreement also explicitly states that ChatGPT will be trained with content from Axel Springers publications.

Publications by Axel Springer have been criticized for major political bias, fake news, sensationalism and political influence. 36% of Axel Springer SE belongs to controversial Investment Company KKR, one of the biggest investors into fossil fuels worldwide.

Quotes

Quote

Mathias Döpfner, chief executive of Axel Springer, said the deal was “the first of its kind”. He added: “We want to explore the opportunities of AI-empowered journalism — to bring quality, societal relevance and the business model of journalism to the next level.” (Financial Times)

 

Quote

The content will get a “favorable position” in ChatGPT search results, with the goal of helping to drive traffic and subscription revenue to Axel Springer brands, according to a source familiar with the deal. (Reuters)

 

My thoughts

As a German, hearing this news made every possible Alarm Bell in my head ring. Axel Springer is a main source of polarization in Germany. Publications by Axel Springer received 41% of all condemnations by the German Press Council in 2023 (30 in total, Bild alone 26). Last year, there was a major controversy about CEO Mathias Döpfner. Private Messages by him were leaked that contained statements like:
 

Quote

"The Ossis (insulting word for people living in former East Germany) are either communists or fascists. They don’t do in-between. Disgusting." Mathias Döpfner

Quote

"Environmental policy – I am very much in favor of climate change. Phases of warmth in civilization have always been more successful than those of cold." Mathias Döpfner

Quote

The leaked messages also raise specific questions about close links between the Springer publishing empire – whose flagship titles includes Bild, Die Welt, Business Insider and Politico – and the pro-business Free Democratic party (FDP), a junior partner in Olaf Scholz’s three-party coalition government.

Citing a dinner with the FDP leader, Christian Lindner, Döpfner repeatedly urged Reichelt as Bild editor to “do more for the FDP” in the run-up to the September 2021 federal elections. “Please strengthen the FDP,” he wrote two days before the vote. “If they do well they can act with such authority in the traffic light [coalition of Social Democrats, Green party and FDP] that it collapses.” [The Guardian]

If ChatGPT starts getting trained with content produced by Axel Springer properties, especially opinion pieces, ChatGPT will turn into a mass spreader of questionable journalism.

Many German Twitter/X Users reacted in shock underneath the Announcement by OpenAI.

Sources

https://www.axelspringer.com/en/ax-press-release/axel-springer-and-openai-partner-to-deepen-beneficial-use-of-ai-in-journalism

https://www.ft.com/content/7cd439bc-29cd-44f9-8676-4761e27bc3a8

https://openai.com/blog/axel-springer-partnership

https://www.reuters.com/business/media-telecom/global-news-publisher-axel-springer-partners-with-openai-landmark-deal-2023-12-13/

https://www.theguardian.com/world/2023/apr/13/axel-springer-ceo-mathias-dopfner-leaked-messages-reported

 

 

[thevictor390]

Unbiased AI is already impossible. What we are calling "AI" is nothing more than a simulation of human written text. It is exactly as biased as the humans it is simulating. Which humans are those? Whoever makes the AI get to decide. Even if you train the AI randomly or on everything, it retains whatever biases are inherent in all humans.

[Taf the Ghost]

All the News that's fit to generate today.

[Skipple]

24 minutes ago, thevictor390 said:

Unbiased AI is already impossible. What we are calling "AI" is nothing more than a simulation of human written text.

This is what people don't understand when they start screaming about bias in AI. Everything has bias. There are very few things in the world that are completely objective. Math, physics, and that's about it. (even physics to an extent is subjective when it comes to relativity) There are very few other topics that are not subjective in nature. 

[SorryBella]

1 minute ago, Skipple said:

There are very few other topics that are not subjective in nature. 

Relevant.

[TempestCatto]

Nah fam. I only get my news from a truly reliable and unbiased source. Memes

[Kisai]

42 minutes ago, thevictor390 said:

Unbiased AI is already impossible. What we are calling "AI" is nothing more than a simulation of human written text. It is exactly as biased as the humans it is simulating. Which humans are those? Whoever makes the AI get to decide. Even if you train the AI randomly or on everything, it retains whatever biases are inherent in all humans.

It will reflect a bias based on it's data ingress. It's absolutely possible to tell trainers to "double weight" a certain source. This you can enforce a bias by telling it to train against certain data multiple times so it becomes a larger part of the training data than otherwise would happen.

 

The bias presently shown, is essentially based on how often certain exact phrases are repeated, word-for-word. This is why data from Wikpedia, Reddit, Quora, StackExchange, etc will overweight in training because people plagiarize, word-for-word, code-for-code from these sources already.

 

So if OpenAI says "train on politco's data 10 times for every 1 time you train on msnbc and wikipedia" then it will learn that bias.

[thevictor390]

6 minutes ago, Kisai said:

It will reflect a bias based on it's data ingress. It's absolutely possible to tell trainers to "double weight" a certain source. This you can enforce a bias by telling it to train against certain data multiple times so it becomes a larger part of the training data than otherwise would happen.

 

The bias presently shown, is essentially based on how often certain exact phrases are repeated, word-for-word. This is why data from Wikpedia, Reddit, Quora, StackExchange, etc will overweight in training because people plagiarize, word-for-word, code-for-code from these sources already.

 

So if OpenAI says "train on politco's data 10 times for every 1 time you train on msnbc and wikipedia" then it will learn that bias.

That's what I mean by "whoever makes the AI gets to decide."

[Lunar River]

Given how much OpenAI like to censor their models (oh the irony), I can't only imagine it is to train in the negative, basically telling it "learn this.. and never say any of it"

[HenrySalayne]

Who honestly needs AI-enhanced tabloid journalism? Or tabloid-journalism-enhanced AI?

 

Axel Springer is already undermining the truth and spreading misinformation - something sadly ChatGPT is also known for, but which did it unintentionally so far. Combining these two is just a further erosion of facts and accelerates the transition into a post-factual world.

[Lunar River]

50 minutes ago, HenrySalayne said:

Who honestly needs tabloid journalism?

Ftfy

[Mark Kaine]

9 hours ago, SorryBella said:

Relevant.

no. musk is never relevant to anything 

[Mark Kaine]

10 hours ago, DerGottesgleiche said:

Summary

gotta say, while i appreciate the effort, you should use google translation for these german news articles,  cause it'll typically do a much better job.  i guess its still legible, but you can't just translate stuff 1:1, its often misleading or simply not clear, in other words, its extremely painful to read and if this was a school test 100% a "6" (F for our American friends lol)

 

 

9 hours ago, Skipple said:

Math, physics

meh, no, absolutely not... like 90% of "theories" are either completely wrong or can't really be proven,  its more like lots of esotheric mambo jumbo with some voodoo sprinkled in rather than actual science... that's not to say the other ~10% aren't right , as we use them daily and they work (mostly)

 

ironically "science" is a lot like Bild zeitung, lots of half thruths, sensationalism and outright lies. 

 

 

very good video on the subject:

 

 

 

 

[SorryBella]

2 hours ago, Mark Kaine said:

no. musk is never relevant to anything 

Yeah, but the grift is so fucking funny to see as Grok becomes "woke" in the face of his "fans". Amazing meltdown to watch in my eyes.

[Skipple]

1 hour ago, Mark Kaine said:

meh, no, absolutely not... like 90% of "theories" are either completely wrong or can't really be proven,  its more like lots of esotheric mambo jumbo with some voodoo sprinkled in rather than actual science... that's not to say the other ~10% aren't right , as we use them daily and they work (mostly)

 

ironically "science" is a lot like Bild zeitung, lots of half thruths, sensationalism and outright lies. 

My point is not that human understanding of math and science is 100% correct, accurate, or communicated well... Simply that there are very few universal truths in... the universe. Everything else is human conceptualization. 

[DerGottesgleiche]

7 hours ago, Mark Kaine said:

gotta say, while i appreciate the effort, you should use google translation for these german news articles,  cause it'll typically do a much better job.  i guess its still legible, but you can't just translate stuff 1:1, its often misleading or simply not clear, in other words, its extremely painful to read and if this was a school test 100% a "6" (F for our American friends lol)

 

 

meh, no, absolutely not... like 90% of "theories" are either completely wrong or can't really be proven,  its more like lots of esotheric mambo jumbo with some voodoo sprinkled in rather than actual science... that's not to say the other ~10% aren't right , as we use them daily and they work (mostly)

 

ironically "science" is a lot like Bild zeitung, lots of half thruths, sensationalism and outright lies. 

 

 

very good video on the subject:

 

 

 

 

Lol, I wrote that Summary myself and didn't translate any German News Article. Don't be too dramatic bro. Not everyone has 6 hours to write a Pulitzer Price worthy news update post on the LTT Forum. Also, using Google Translate is by now probably the worst choice for translating something. ChatGPT or Deepl are way better at translating.

[mr moose]

8 hours ago, Mark Kaine said:

. like 90% of "theories" are either completely wrong or can't really be proven,  its more like lots of esotheric mambo jumbo with some voodoo sprinkled in rather than actual science... that's not to say the other ~10% aren't right ,

 

 

 

eleventytwo percent of online stats are correct twelveteen times out of 10.

[Mark Kaine]

2 hours ago, DerGottesgleiche said:

ChatGPT

ah, that explains the terrible English/ German translation in these news "articles" it literally doesn't sound like a human wrote them, don't take it personally, if you didn't write/translate them