🗞️In this edition

• Federal court compels ChatGPT log disclosure in copyright case

• Sponsored: SERPView - One dashboard for all your properties

• Anthropic CEO acknowledges billions in data center uncertainty

• Huang tells Trump China won't accept "degraded chips”

• In other AI news –

  • Amazon’s Trainium2 already a billion dollar heavyweight

  • Wordpress Telex starts proving itself in real workflows

  • Neptune joins Openai to strengthen training tools

  • 4 must-try AI tools

Judge orders OpenAI to disclose 20 million ChatGPT logs

What's happening:

A federal judge in Manhattan ordered OpenAI to disclose millions of anonymized chat logs from ChatGPT platform. The decision, handed down by the U.S. Magistrate Judge Ona Wang, stems from a copyright infringement lawsuit brought by New York Times and other news organizations.

Lawsuit, initiated in 2023, accuses OpenAI and partner Microsoft of unlawfully using copyrighted material to train AI models. Plaintiffs argue ChatGPT reproduces and distorts their articles without permission, effectively siphoning value from original journalism.

OpenAI fought to shield user interaction data, claiming revealing logs would compromise trade secrets and user privacy. Judge Wang rejected these arguments, ruling anonymized logs essential for plaintiffs to prove claims of direct copying.

Ordered disclosure involves up to 20 million chat logs, a massive trove that could reveal how often ChatGPT regurgitates protected content. Logs must be produced in de-identified format, stripping personal identifiers while preserving substance of user queries and AI responses.

At the heart of the dispute is the question of fair use in AI training. OpenAI contends ingesting vast datasets from the internet falls under fair use doctrines, transforming raw data into innovative tools. Plaintiffs counter that this process involves verbatim reproduction, not mere transformation, violating copyright laws.

This case is part of a larger offensive by media outlets against AI giants. Over 60 copyright suits filed against AI firms in the U.S., with New York Daily News, U.S. News, and others joining. In Canada, Ontario court allowed a parallel suit by news publishers to proceed.

Why this is important:

20 million chat logs revealing how often ChatGPT reproduces copyrighted content could establish a pattern of infringement central to plaintiffs' case.

OpenAI's repeated failures to halt such orders suggest courts prioritizing accountability over corporate secrecy. This is the third major disclosure order in copyright cases against AI companies.

The anonymization requirement addresses privacy concerns but experts worry even de-identified data could be reverse-engineered. Anything users said to ChatGPT might be exposed in legal discovery.

Over 60 copyright suits against AI firms with financial stakes estimated in billions shows systematic legal pressure forcing industry toward paid data partnerships.

Our personal take on it at OpenTools:

This is a discovery process forcing transparency OpenAI desperately wants to avoid.

20 million chat logs is massive exposure. If analysis shows ChatGPT frequently regurgitating news articles verbatim, that's direct evidence of infringement undermining fair use defense.

But OpenAI created this problem by training on copyrighted content without permission then claiming fair use. Discovery is a consequence of that choice.

OpenAI's recent content deals with select publishers are attempt to moot litigation but too little, too late. Can't retroactively license training data after you already used it.

The broader implication is every AI company training on web data faces similar exposure. If NYT wins on evidence from these logs, 59 other plaintiffs have a roadmap to victory.

A German court already found OpenAI liable for reproducing song lyrics. A Canadian court allowed the news publisher's suit to proceed. US courts ordering log disclosure. OpenAI's losing on multiple fronts simultaneously.

Appeals are likely but OpenAI's track record on appeals in these cases is poor. Courts consistently rule against them on discovery.

Read More

Manage All Your Search Console Properties in One Place

Google Search Console is kind of a mess. You can only see 1,000 rows of data (when there are often 50,000+ queries). Switching between client accounts is annoying. And good luck finding which keywords are actually worth your time.

SERPView fixes this:

• See all 50,000+ rows of your search data, not just 1,000

• One dashboard for all your sites (no more logging in and out)

• Shows you "striking distance" keywords (you're ranking #11-20, easy wins), and even more reports which you can even think of. 

• Catches keyword cannibalization before it kills your rankings

It's basically GSC but usable. They have a lifetime deal going right now. Code PH10 gets you another 10% off.

Try SerpView today

Anthropic CEO warns AI infrastructure spending carries major risk

What's happening:

Dario Amodei, CEO of Anthropic, acknowledged Wednesday that the AI industry is taking considerable risk as it spends hundreds of billions on data centers powering AI.

"On the economic side, I have concerns that even if technology is really powerful and fulfills all promises, I think with some players in the ecosystem, if they get it off by little bit, bad things could happen," he said at DealBook Summit.

Amodei, whose company is among those spending billions on new computing facilities, declined to say which companies might find themselves in a difficult position. But he acknowledged even his own company faces uncertainty building data centers that won't be finished for years.

"It's a real dilemma with how economic value will grow and lag time on data centers that drives it," Amodei said. "I think it's a genuine uncertainty and a genuine dilemma."

In September, Anthropic completed a funding round valuing it at $183B, up nearly three times from $61.5B six months earlier. OpenAI is now worth an estimated $500B.

But today, AI technologies are often losing money. Almost 8 in 10 businesses have started using generative AI, but just as many said it has had "no significant bottom-line impact," according to McKinsey research. Anthropic is not profitable. Neither is OpenAI.

Why this is important:

Anthropic CEO publicly admitting "bad things could happen" if companies "get it off by little bit" is rare candor about AI infrastructure risk.

80% of businesses using generative AI reporting no significant bottom-line impact while companies spend billions validates concerns about ROI mismatch.

Neither Anthropic ($183B valuation) nor OpenAI ($500B valuation) is profitable despite massive valuations. That's unsustainable without revenue growth materializing.

"Cone of uncertainty" where CEOs must decide now on 2027 compute needs shows multi-year infrastructure commitments happening before demand is proven.

Anthropic investing $50B in data centers while CEO warns of timing risk shows even conservative players can't escape dilemma.

Our personal take on it at OpenTools:

This is Anthropic's CEO saying the quiet part out loud.

"If they get it off by little bit, bad things could happen" is an admission that AI infrastructure spending is a massive bet where small miscalculations create catastrophic outcomes.

80% of businesses reporting no bottom-line impact from AI is damning stat. Enterprise adoption isn't translating to measurable value yet. If that continues, demand projections underpinning infrastructure spending collapse.

The multi-year lag between infrastructure decisions and revenue realization creates fundamental uncertainty. You're betting billions on 2027 demand based on 2024 trends. What if trends change?

This is different from past infrastructure buildouts. Telecom overbuilt fiber in late 1990s but fiber was the utility everyone needed eventually. AI compute might not be. If AI plateau before matching hype, stranded capital is massive.

Amodei's candor is refreshing but also alarming. If most conservative AI CEOs are publicly worried about timing risk, how exposed are aggressive players?

Read More

Huang doubts China wants H200

What's happening:

Nvidia CEO Jensen Huang said he's unsure whether China would accept H200 AI chips if US restrictions are relaxed, following a Wednesday meeting with President Trump. Asked if Beijing would allow Chinese companies to buy H200, Huang said, "We don't know. We have no clue. We can't degrade chips that we sell to China, they won't accept that."

Trump administration officials discussed whether to allow H200 sales to China. Allowing sales would mark a significant win for Nvidia, which has pressed for export control relaxation to access the world's second-largest economy.

Huang forged a close relationship with Trump since the November election, arguing restrictions only boost China's domestic champions like Huawei. Asked how often he's in Washington, Huang said: "Whenever President Trump would like me to be here."

Nvidia secured lobbying win by keeping Gain AI Act out of defense legislation. The Act would require chipmakers to give American customers first dibs on powerful AI chips before selling to China.

China told domestic customers to shun Nvidia's H20 chip last summer, favoring Chinese processors instead. Efforts to export hobbled Blackwell-generation chips failed during October's Trump-Xi meeting.

Huang said China represents a $50B market for Nvidia, though data center revenue from Asia is excluded from forecasts.

Why this is important:

China telling customers to shun H20 last summer and Blackwell efforts failing in October show Beijing already rejecting Nvidia's export-compliant chips in favor of Huawei and domestic players.

$50B market opportunity versus uncertain acceptance creates tension: Nvidia wants access but China may not want restricted chips even if the US allows sales.

Our personal take on it at OpenTools:

Huang's lobbying Trump for H200 export approval while admitting "we have no clue" if China wants them is a bizarre position.

China already rejected H20 (designed to fall below export limits) by telling customers to use domestic chips instead. Why would they accept H200, another export-restricted chip, after snubbing H20? They won't.

Beijing's strategy is clear: use US export controls as justification to force adoption of Huawei and domestic alternatives. Accepting Nvidia chips undermines that narrative. China's choosing self-sufficiency over access to degraded American chips.

Nvidia keeping Gain AI Act out of defense legislation is a real lobbying win. The Act would have required American customers to get priority access before international sales. Blocking it preserves Nvidia's ability to sell globally if restrictions lift.

But the failed Blackwell export attempt during the October Trump-Xi meeting shows even Trump couldn't persuade Xi to accept restricted chips. If Trump can't broker that deal, Huang's lobbying probably won't either.

This is Nvidia trying to have it both ways: lobby for export approval to show shareholders they're pursuing China market, while knowing China likely won't accept restricted chips anyway. That preserves optionality without a realistic path to revenue..

Read More

📢 What else is happening?

• Andy Jassy says Amazon’s Nvidia competitor chip is already a multibillion-dollar business – He said the Trainium2 business “has substantial traction, is a multi-billion-dollar revenue run-rate business, has 1M+ chips in production, and 100K+ companies using it as the majority of Bedrock usage today.”

• WordPress’s vibe-coding experiment, Telex, has already been put to real-world use – Telex had been used within a working WordPress shop to do things like create price comparisons, price calculators, and pull in real-time business hours plus a map link to a retail store, among other examples.

• OpenAI to acquire Neptune, a startup that helps with AI model training – Neptune builds monitoring and de-bugging tools that AI companies use as they train models.

👩🏼‍🚒Discover mind-blowing AI tools

1. ReplaiGPT - An AI-powered email reply tool that generates personalized responses using pre-defined context

2. Vsub - A service that offers video captioning and subtitle generation

3. BlueWillow - An AI tool designed to help users create logos, graphics, photo-realistic scenes

4. CoverLetterWrite - Helps users write personalized cover letters quickly