Lightnews — Scholar-powered news

Reposted by Ed Newton-Rex

Molly White @molly.wiki · 8d

Someone should probably inform the White House's "AI & Crypto Czar" that no one is forcing AI companies to train their models on Wikipedia

Tweet by David Sacks: "Wikipedia is hopelessly biased. An army of left-wing activists maintain the bios and fight reasonable corrections. Magnifying the problem, Wikipedia often appears first in Google search results, and now it’s a trusted source for AI model training. This is a huge problem."

100 580 3.2K

Ed Newton-Rex @ednewtonrex.bsky.social · 8d

More good news on AI & copyright:

OpenAI reportedly on track to lose in German copyright case over song lyrics.

aifray.com/breaking-ope...

16 51

Ed Newton-Rex @ednewtonrex.bsky.social · 8d

Genuinely emotional reading this. $1.5 billion to authors, in the biggest copyright settlement in history.

Big tech is not above the law. This is just the start.

www.nytimes.com/2025/09/29/o...

10 35

Ed Newton-Rex @ednewtonrex.bsky.social · 9d

The Tony Blair Institute report: institute.global/insights/tec...

What the UK Thinks About AI: Building Public Trust to Accelerate Adoption

institute.global

3

Ed Newton-Rex @ednewtonrex.bsky.social · 9d

When you run a poll and find that the public is concerned about AI, it takes crazy levels of AI boosterism to conclude that the public is wrong and the govt must educate them.

Perhaps… the public is right.

Politicians should listen to the public, not the Larry Ellison-funded Tony Blair Institute.

2 9 39

Reposted by Ed Newton-Rex

The Guardian @theguardian.com · 13d

Adviser to UK minister claimed AI firms would never have to compensate creatives

Exclusive: Kirsty Innes made statement in now-deleted post on X seven months before taking up role as Liz Kendall aide * UK politics live – latest updates A senior ministerial aide said AI companies will never have to compensate creatives for using their content to train their systems, in a statement that has alarmed campaigners demanding Labour delivers a fairer deal for musicians, artists and writers from the tech industry. Kirsty Innes, recently appointed as a special adviser to Liz Kendall, the secretary of state for science, innovation and technology, said “whether or not you philosophically believe the big AI firms should compensate content creators, they in practice will never legally have to”. Continue reading...

www.theguardian.com

12 28 39

Ed Newton-Rex @ednewtonrex.bsky.social · 13d

It is very worrying that Kendall is bringing in an adviser who has

- said AI companies will never compensate creatives
- echoed big tech’s calls for copyright reform

Why is this government so set on throwing creatives under the bus to benefit AI companies?

www.theguardian.com/technology/2...

1 3 15

Ed Newton-Rex @ednewtonrex.bsky.social · 14d

The UK government’s new tech adviser has endorsed changing copyright law to give AI companies free access to copyrighted work, with no opt-out for creators.

Labour is the party of big tech.

2 14 25

Ed Newton-Rex @ednewtonrex.bsky.social · 14d

To her credit, she invited people to respond; I did. I desperately hope her views have changed.

/end

6

Ed Newton-Rex @ednewtonrex.bsky.social · 14d

In her thread in February, she said big AI firms "in practice will never legally have to [compensate content creators]", and concluded that we need to "facilitate model training" to "maximis[e] UK growth".

2/3

1 6

Ed Newton-Rex @ednewtonrex.bsky.social · 14d

It is concerning to hear that Kirsty Innes has been brought in as a special adviser at DSIT, given her previous comments on AI & copyright.

(Her February thread on AI & copyright, and my response, in screenshots ⬇️)

🧵 1/3

2 4 14

Ed Newton-Rex @ednewtonrex.bsky.social · 22d

In each case, the logic is: I won’t stop this by resisting, since presumably others will take the deal. So why turn it down?

The only way to escape the model trainer’s dilemma is rapid, large-scale organisation. It remains to be seen whether that’s possible in the timeframe required.

/end

6

Ed Newton-Rex @ednewtonrex.bsky.social · 22d

3. Domain experts training models at the reinforcement stage. They are training a model to do their job, and get a tiny hourly wage to do so - but if they don’t, someone else will.

5/6

1 2

Ed Newton-Rex @ednewtonrex.bsky.social · 22d

2. Individual creatives. They get a bad offer from the media company that represents them, giving them little money and less control, in exchange for the right to license their work to AI companies. But they think they may as well take the deal, as others no doubt will.

4/6

1 2

Ed Newton-Rex @ednewtonrex.bsky.social · 22d

1. Publishers / media companies. They are offered a nice-looking sum to license their work - but it’s a perpetual license, so they’ll only see the revenue once.

The deal is bad, but they figure it may as well be them who makes this money, as otherwise it will be their competitors.

3/6

1 2

Ed Newton-Rex @ednewtonrex.bsky.social · 22d

… - but, without knowing how others will act, each is incentivized to be the one to take the deal.

Seeing it in at least 3 groups:

2/6

1 2

Ed Newton-Rex @ednewtonrex.bsky.social · 22d

The model trainer’s dilemma is a specific instance of the prisoner’s dilemma, and it’s everywhere in AI right now.

AI companies are offering people bad deals to help train their own replacements. These people would collectively be better off if they resisted…

🧵 1/6

2 1 9

Ed Newton-Rex @ednewtonrex.bsky.social · 23d

AI company CEOs say they want to solve humanity’s hardest problems, and then are like ‘try our micro blueberry model to generate images of yourself as anime lol, by the way it’s $20/month’

8 38

Ed Newton-Rex @ednewtonrex.bsky.social · 29d

I went to see Michael Trazzi on day 4 of his hunger strike at Google DeepMind

There’s the beginnings of a broad alliance taking shape - different groups, voicing different concerns about AI, but all aligned in a belief that AI companies must stop racing ahead and ignoring the harm they are causing.

1 1 15

Ed Newton-Rex @ednewtonrex.bsky.social · Sep 7

Anthropic have said publicly that they use synthetic data, and have even said they think they can build an “infinite data generation engine”.

The settlement gets some money to the authors of the pirated books (good), but fails to stop Anthropic benefiting from them via synthetic data (very bad).

2 6 22

Ed Newton-Rex @ednewtonrex.bsky.social · Sep 6

Still, $1.5B into the pockets of authors is a big win for creatives, and is why so many people fight so hard for creatives to be treated fairly by AI companies. And I suspect we may see even bigger sums from other cases where copyright owners win outright.

1 1 15

Ed Newton-Rex @ednewtonrex.bsky.social · Sep 6

But it’s frustrating that big AI companies can use the hype around AI to settle these lawsuits so easily. $1.5B is *3x* how much Anthropic was worth when they downloaded these pirated books in 2021; now, $1.5B is less than 1% of their value.

2/3

1 14

Ed Newton-Rex @ednewtonrex.bsky.social · Sep 6

The $1.5 billion settlement in the Anthropic copyright lawsuit is historic - the biggest ever copyright case payout, $3k each for 500,000 authors.

🧵 1/3

1 2 16

Ed Newton-Rex @ednewtonrex.bsky.social · Sep 5

So. Farewell
Then
Peter Kyle.

You said
Copyright law was
Uncertain.

And then
You were uncertain
About what
You had said.

You are certainly
No longer
Technology minister.

EJ Thribb

1 2 16

Ed Newton-Rex @ednewtonrex.bsky.social · Aug 29

... and on those in government who succumb to the millions these companies spend on lobbying, and who cheer them on.

But the public sees the costs of AI. And it seems pretty clear that, if AI companies continue down the path they're on, the backlash will only continue to grow.

/end

4 25