Anthropic forced to pay $1.5B in AI copyright settlement; OpenAI wants to help get you hired
Get bigger weekly updates! Free subscribers receive the top stories each week, while Paid subscribers will get a few extra stories. All support for Handy AI directly helps me maintain the newsletter and keep the information flowing.
last week’s top stories
⚖️ Anthropic to pay $1.5B in AI copyright settlement. Anthropic has agreed to pay at least $1.5 billion to settle a class-action lawsuit by authors who accused the AI startup of using pirated books to train its Claude chatbot without permission. The landmark deal (about $3,000 per infringed book) requires Anthropic to destroy the illicit data and marks the first major copyright settlement of the AI era, sending a message that AI firms must compensate creators for training material. Read more
💼 OpenAI launching AI-powered job platform. OpenAI announced plans for an AI-driven hiring platform to connect companies with skilled workers, putting it on a collision course with LinkedIn. The “OpenAI Jobs Platform,” slated for mid-2026, will use AI matching to pair candidates with roles (including a track for small businesses and local governments) and will offer certification programs through an OpenAI Academy to verify people’s AI skills. It’s part of OpenAI’s expansion beyond ChatGPT into new markets amid concerns about AI-driven job disruption. Read more
🏛️ Tech giants pledge AI education initiatives at White House. In a high-profile White House dinner, President Trump hosted CEOs including Mark Zuckerberg, Sam Altman, Tim Cook, Sundar Pichai and others to announce major investments in AI education. Microsoft will provide U.S. college students with free access to Office 365 Copilot and fund AI training grants for teachers; Google pledged $150 million for AI job skills programs; Amazon announced new AI courses for millions of learners; and Anthropic committed $1 million toward K-12 AI curricula and cybersecurity training. The gathering underscored a national push to equip America’s workforce with AI skills, with tech leaders citing hundreds of billions in planned U.S. tech investments. Read more
💻 OpenAI plans custom AI chips with Broadcom. OpenAI is partnering with Broadcom to design its own AI acceleration chips, aiming to reduce reliance on Nvidia GPUs. The first in-house chip is expected in 2026 and will be used internally to power ChatGPT and other models. This follows similar moves by Google, Amazon, and Meta to build bespoke silicon as AI computing demands skyrocket. Read more
🤖 DeepSeek prepping self-improving AI agent. Chinese startup DeepSeek is developing an autonomous AI agent designed to perform multi-step tasks with minimal human guidance and learn from its own actions. Founder Liang Wenfeng is pushing to unveil the agent in Q4 2025, following DeepSeek’s earlier “R1” model that surprised the industry with its low-cost, high-performance approach. This move joins a broader industry push (by OpenAI, Anthropic, Microsoft, etc.) toward AI “agents” that can execute complex workflows, though DeepSeek candidly notes challenges like AI “hallucinations” remain unresolved. Read more
🍎 Apple rumored to tap Google’s AI for Siri upgrade. A Bloomberg report claims Apple is in talks with Google to power a revamped Siri using Google’s forthcoming Gemini AI model. Under a possible agreement, Apple would test a custom version of Google’s AI to significantly boost Siri’s capabilities by 2026, potentially integrating advanced web search, AI summaries, and voice-controlled device navigation. If true, the move would be a surprising collaboration between rivals and an attempt by Apple to catch up in the AI assistant race by leveraging Google’s tech. Read more
💰 Anthropic raises $13B at $183B valuation. Anthropic has closed a massive $13 billion Series F funding round that values the AI lab at approximately $183 billion post-money. The round was led by Iconiq Capital with participation from major VCs, sovereign wealth funds, and big investors like BlackRock and Google’s fund. Anthropic says the cash will fuel enterprise adoption of its Claude model, safety research, and global expansion – reflecting surging demand (its annual revenue run-rate jumped from $1B to $5B this year). The monster raise triples Anthropic’s valuation since March and underscores the feverish investor confidence in top-tier AI startups. Read more
🤝 OpenAI acquiring Statsig in $1.1B deal. OpenAI announced it will acquire Statsig, a feature-testing and experimentation startup, in an all-stock deal valuing Statsig at about $1.1 billion. As part of the deal, Statsig’s CEO Vijaye Raji will become OpenAI’s “CTO of Applications,” overseeing product engineering for ChatGPT and other consumer-facing tools. The acquisition, following OpenAI’s $6.5B purchase of Jony Ive’s hardware startup last quarter, bolsters OpenAI’s capabilities in rapid product iteration (Statsig built tools to A/B test and roll out features) as it scales up ChatGPT amid intense competition. Read more
📚 Apple sued over allegedly training AI on pirated books. Apple has been hit with a proposed class-action lawsuit by two authors who claim the company used a trove of pirated e-books (the Books3 dataset) to train its AI models. The writers allege Apple’s secretive “OpenELM” language model was built from copyrighted novels (including their own works) scraped without permission or pay – essentially “stealing books” to fuel AI. The lawsuit, filed in California, comes amid a wave of similar cases (OpenAI, Meta, and others face suits too) as authors fight back against unlicensed use of their content in AI training. Read more
💸 OpenAI projected to burn $115B by 2029 on AI development. OpenAI has dramatically increased its spending forecasts, now expecting to burn through $115 billion in cash by 2029 to fuel its AI ambitions. According to The Information, the startup is on track to spend over $8B this year alone (double earlier estimates) and more than $17B next year, driven by enormous cloud-compute needs for training and running models like ChatGPT. Read more
🧪 AI Research of the Week
Why Language Models Hallucinate
From OpenAI
Jake's Take: OpenAI argues hallucinations arise because today’s training and leaderboard scoring reward guessing over expressing uncertainty. Their new paper formalizes how pretraining likely encourages shaky facts and shows post-training incentives are encouraging models to always provide an answer (even if its a wrong one).
The proposed fix is simple, in theory: change primary evals to penalize confident errors and give credit for “I don’t know.” On their blog, OpenAI contrasts two models with similar accuracy but radically different error rates, highlighting how accuracy-only scoreboards hide hallucinations.
If you’re build AI apps, treat your model’s abstention (those “I don’t knows”) as a first-class output, encourage more calibrated confidence, and align reinforcement learning rewards with uncertainty-aware scoring.
and then, even more news…
🕵️ Musk’s xAI sues ex-engineer for trade secret theft. Elon Musk’s AI startup xAI filed a lawsuit accusing a former researcher, Xuechen Li, of stealing confidential data on his way out to join OpenAI. The suit alleges Li copied “cutting-edge” internal documents about xAI’s Grok model onto personal devices right after cashing out ~$7M of xAI stock, then tried to cover his tracks by deleting logs and renaming files. xAI claims the stolen secrets (which Li reportedly admitted taking) could save competitors “billions” in R&D. Read more
📁 ChatGPT’s Projects feature now free for all. OpenAI has opened up Projects in ChatGPT to all users, including those on the free tier. Projects let you create dedicated workspaces with separate memory and files for specific tasks (e.g. writing or coding projects) instead of everything living in one chat thread. The update means free users can now organize multiple conversations and uploads per project. OpenAI also added a “branch in new chat” option to fork off side conversations without losing context, making it easier to explore tangents without derailing your main thread. Read more