Hi there, this is your daily ☕️ Techpresso.
In today's Techpresso:
💥 Anthropic to pay $1.5B to settle pirated books case
💰 EU fines Google $3.5 billion for anti-competitive ad practices
🏛️ Nvidia opposes a bill to keep GPUs in America
💸 OpenAI reportedly misjudged cash burn by $80 billion
📚 Apple sued for using pirated books for AI
🎁 + 9 other news you might like
🔮 + 5 handpicked research papers and tools
💥 Anthropic to pay $1.5B to settle pirated books case LINK
Anthropic agreed to a historic $1.5 billion settlement that will give approximately half a million authors a minimum payment of $3,000 each for the unauthorized use of their work.
The lawsuit centered on the company pirating millions of books from “shadow libraries” to ingest data into its LLMs, not on the actual act of training its AI system, Claude.
A judge previously ruled that training AI on copyrighted material is “transformative” under the fair use doctrine, so this specific settlement only addresses the act of illegal downloading.
💰 EU fines Google $3.5 billion for anti-competitive ad practices LINK
The European Commission is fining Google $3.5 billion for distorting advertising competition by giving its own ad tech software preferential treatment over rivals in the online marketplace.
The EU found that Google informed its AdX exchange of the best rival bid in advance, while its Google Ads tool also avoided competing exchanges to place bids mainly on AdX.
Google now has 60 days to address the Commission's issues or it could face an "appropriate remedy," which might include being forced to sell off parts of its adtech business.
🏛️ Nvidia opposes a bill to keep GPUs in America LINK
Nvidia is publicly opposing the "GAIN AI Act," a proposed law that would force the company to sell its leading GPUs to American buyers first before they could be exported abroad.
The legislation prevents GPU exports if a "current backlog of requests from United States persons" exists, potentially restricting more products than are currently banned from sale in China.
In a public statement, the chipmaker argued the bill tries to "solve a problem that does not exist" and would hamper global competition in any industry that uses mainstream computing chips.
💸 OpenAI reportedly misjudged cash burn by $80 billion LINK
OpenAI now forecasts a total cash outflow of $115 billion through 2029, which is about $80 billion more than earlier projections that had the company reaching break-even.
The soaring costs are mainly due to expensive computing power and a plan to invest almost $100 billion in data centers and custom chips to reduce cloud provider dependency.
To cover new expenses, the company projects $200 billion in revenue by 2030 and plans to generate about $110 billion from free ChatGPT users through commissions or ads.
📚 Apple sued for using pirated books for AI LINK
A new proposed class action lawsuit accuses Apple of training its `OpenELM` AI model on pirated books, citing the company’s own technical paper as direct evidence for the claim.
The suit explains that Apple's paper names the `RedPajama` dataset for training, which includes the `Books3` dataset, a source the lawsuit calls a known body of pirated books.
The authors request the court allow a class action and seek statutory damages of $150,000 per infringed work, plus a restitution of profits following a jury trial.
Other news you might like
- Trump threatens trade probe after 'discriminatory' EU fines against Google, AppleLINK
- Zuckerberg caught on hot mic telling Trump ‘I wasn’t sure’ how much to promise to spend on AI in the USLINK
- OpenAI reorganizes research team behind ChatGPT’s personalityLINK
- Roblox announces short-form video feed for gameplay clips, new AI tools for creators, and moreLINK
- Qualcomm CEO says Intel ‘not an option’ for chip production — yetLINK
- Microsoft’s light-powered computer could run AI 100x faster and more efficientlyLINK
- AppLovin and Robinhood added to S&P 500LINK
- Google leads monster week for tech, pushing megacaps to combined $21 trillion in market capLINK
- Personalized AI companion app Dot is shutting downLINK
Latest research and tools
Qwen3 30B: an AI model that runs at 13 tokens per second on a cluster of four Raspberry Pi 5 computers.LINK
Fantastic pretraining optimizers and where to find them: this paper finds that a carefully tuned AdamW optimizer is a very strong and reliable choice for pretraining large language models, often outperforming newer alternatives.LINK
PlateShapez: a research tool that generates modified license plate patterns on vehicle images to test the reliability of automated reader systems.LINK
A Software Development Methodology for Disciplined LLM Collaboration: a structured approach for developers working with AI that uses systematic constraints to prevent common issues like code bloat and architectural drift.LINK
Contemplative Artificial Intelligence: the paper proposes an AI that improves its reasoning by pausing to think and self-critique its own answers before responding.LINK
Want to get the latest news differently? Find us on:
See you tomorrow for a new dose of ☕️ Techpresso!
.png){kind=small}
.png){kind=small}
.png){kind=small}