- The Shift
- Posts
- The Biggest AI Copyright Settlement Yet
The Biggest AI Copyright Settlement Yet
Plus, 🧠 Turn Product Photos into Talking Avatars for Ads or Landing Pages, Alibaba’s trillion-parameter model, and more!
Hello Readers👀
Curious about the biggest AI moves today? You’re in the right place. In today’s edition, we have:
📚 Anthropic Pays $1.5B to Settle Landmark AI Copyright Case
🧠 Turn Product Photos into Talking Avatars for Ads or Landing Pages
🚀 Qwen3-Max-Preview: Alibaba’s 1T-Parameter Model
🔨Tools and Shifts you cannot miss
📚 Anthropic Pays $1.5B to Settle Landmark AI Copyright Case
Anthropic has agreed to a $1.5 billion settlement after being accused of training Claude on millions of pirated books. The case, which centered on shadow libraries like LibGen, marks the first major test of how copyright law applies to generative AI. If approved, it will be the largest payout of its kind in U.S. history.
The Shift:
1. The Pirated Dataset Problem - Authors sued after discovering Anthropic ingested over 7 million pirated books from shadow libraries to train Claude. While some books were legally obtained, the lawsuit zeroed in on the massive trove of unlicensed works.
2. Court’s Fair Use Boundary - In June, Judge Alsup drew a sharp line: legally purchased books can be used for AI training under fair use, but pirated copies cannot. That ruling set the stage for a high-stakes trial before the parties struck a deal.
3. Settlement Terms - The agreement covers about 500,000 works at $3,000 each, with additional payouts if more pirated content is uncovered. Anthropic must also destroy all pirated data and backups, and the deal does not authorize future training on such materials.
This case forces AI companies to move toward licensed datasets and compensation schemes rather than scraping shadow libraries. It signals the beginning of a new compliance era, where creative industries can push back and shape how their work fuels AI systems. The ripple effect could influence ongoing cases against Meta, OpenAI, and Midjourney, reshaping how the entire sector handles training data.
🧠 Turn Product Photos into Talking Avatars for Ads or Landing Pages
You can now make any static photo (even a product ambassador) come alive, perfect for DTC founders or performance marketers running UGC-style ads at scale.
🚀 How to Use Aurora (Creatify)
Go to https://creatify.ai
Upload a face photo: (Or choose one of Creatify’s pre-built avatars)
Add your voice
Upload an audio clip or
Type a script (Aurora will generate AI voice + lip sync)
Hit Generate: The avatar now gestures, speaks, and animates, fully synced.
🎯 Why This Matters
Create unlimited ad variants without actors or shoots
Build trust with consistent spokesperson avatars
Repurpose testimonials, founder intros, or product demos
Save time on editing, it’s auto-generated in minutes
🚀 Qwen3-Max-Preview: Alibaba’s 1T-Parameter Model
Alibaba’s Qwen team has launched Qwen3-Max-Preview, its largest LLM yet, with over 1 trillion parameters. Benchmarks show it beating Claude Opus 4, Kimi K2, and DeepSeek-V3.1 across reasoning, coding, and advanced benchmarks, while community testers highlight its blazing-fast response speed.
The Shift:
1. Scale and Benchmarks - Qwen3-Max-Preview scales to 1T parameters, placing it among the largest models in existence. It leads on tests like SuperGPQA, AIME25, LiveCodeBench v6, and Arena-Hard v2, surpassing previous Qwen models and U.S. lab rivals. Early trials show strong reasoning outputs despite not being formally marketed as a reasoning model.
2. Technical Specs and Access - The model supports a 262K token context window with context caching for long sessions. Designed for reasoning, coding, structured data, and creative tasks, it is accessible via Qwen Chat, Alibaba Cloud API, OpenRouter, and AnyCoder.
3. Pricing and Enterprise Considerations - API pricing is tiered by input size, starting at $0.861 per million tokens and rising with longer prompts, making smaller workloads cost-efficient, but heavy enterprise deployments could become expensive. Enterprises gain power and flexibility, but preview status raises questions on stability, compliance, and long-term costs.
Qwen3-Max-Preview shows Alibaba pushing scale while rivals shrink models for efficiency, proving “scaling still works.” Its trillion-parameter milestone cements Qwen as a global contender against U.S. labs, but with costs, closed access, and enterprise risks, adoption will require careful weighing of power versus practicality.
🔨AI Tools for the Shift
🔮 Tarotify – Make tarot reading simple with AI that explains every card instantly, no prior knowledge needed.
📈 Marketbetter – Turn your GTM signals into real meetings with AI-driven precision.
🧾 InvoiceClip – Upload invoices, bills, or receipts and get organized, accurate data instantly without manual typing.
🌍 Glotera AI – Type naturally and translate immediately in any app with seamless AI translation.
📸 PhotoG – Generate ads, videos, and SEO content from just one image.
🚀Quick Shifts
🔍 Google has officially clarified Gemini usage limits: free users get 5 prompts per day with Gemini 2.5 Pro, 100 with AI Pro, and 500 with AI Ultra. Free users also receive 5 Deep Research reports and 100 images daily, while paid tiers expand this to 1,000 images.
🔎 OpenAI is folding its 14-person Model Behavior team into Post Training, signaling personality shaping is now central to model development. Founder Joanne Jang will launch OAI Labs, exploring new AI-human interfaces.
✨ OpenAI’s Joanne Jang, former Model Behavior lead, launched OAI Labs to prototype new AI-human collaboration interfaces, aiming to move beyond chatbots toward creative, interactive systems for thinking, learning, and making.
✍️ Authors have filed a class-action lawsuit against Apple, accusing it of training its OpenELM large language models using a pirated book dataset without permission, credit, or compensation.
🔥 OpenAI’s projected cash burn surged to $115B through 2029, an $80B increase. Annual spending will ramp from $8B in 2025 to $45B in 2028, funded by chip, data center, and compute expansion plans.
Reply