Welcome back, Superhuman. It’s almost time to stop trusting voices. A new model can clone any voice and then use it in a conversation seconds later — with startling accuracy. Meanwhile, over 1,000 employees at the big AI labs signed a joint letter warning that AI is developing faster than our ability to understand and control it.
Today: Why AI companies are after pre-2022 books, how to record a Skill in Claude and reuse it, and get the latest prompts and trending social posts.
TODAY IN AI
Watch how Fish Audio’s new model can clone voices instantly. Source: Fish Audio.
1. Over 1,000 employees at frontier AI companies speak out: A new joint letter warns that AI’s capabilities could accelerate “beyond our ability to understand or control the resulting systems.” The letter urges the US government to deliberately pace development at an international level, which would curb the competitive pressure each lab feels individually. It’s signed by the chief scientists at OpenAI, Anthropic, and Meta, among many others. Read the full letter here.
2. A new model can replicate voices in seconds: Fish Audio publicly launched its leading voice model, S2.1 Pro, with the founders letting it clone their voices in real-time during a product demo. In its first year, the startup built a library of 2M+ community-uploaded voices, grew to $21M in annual recurring revenue, and recently announced $52M in seed funding. It’s run by co-founders Rissa Cao (formerly Amazon Alexa) and Shijia Liao (formerly Nvidia). Try out a voice.
3. Perplexity’s agentic worker goes live on Windows and gets Kimi K3 access: Perplexity Computer can now work with local files on Windows PCs, including across Word, Excel, and PowerPoint. Perplexity also shipped Model Council, a feature that sources answers from multiple LLMs to help surface knowledge gaps in their answers. This includes the recently-released Kimi K3, which is now available on Perplexity for Pro and Max subscribers. Try it here.
Yesterday’s most-clicked story: Moonshot AI released the full weights for its sold-out new model, Kimi K3.
SPONSORED BY SUPERBLOCKS
Recently, a vibe-coded app leaked 1.5M API keys because nobody reviewed the code. Your teams ship AI-written code every day. Superblocks keeps it governed:
Every app is scanned for vulnerabilities before it ships
Runs in your own AWS account, so data never leaves your network
See every app, who built it, and what data it touches, in one place
FROM THE FRONTIER
There’s a hot new commodity for training AI models — your local bookstore
Made with Midjourney
Used booksellers are experiencing a bizarre phenomenon
Bulk orders. Most used bookstores sell roughly one or two dozen books per week. Now some are selling hundreds, if not thousands. The owners have noticed a few trends: buyers are after pre-2022, low-circulation books, often with odd titles. Many are convinced AI companies are behind it, scooping up these books to use them as training data — and Anthropic’s Project Panama is a major reason why.
Anthropic’s plan to digitize “all the books in the world”
Court filings from 2025 revealed that Anthropic purchased and digitized millions of books to train its LLMs, through an internal program called Project Panama. It would buy books, scan them, and often destroy them in the process. A judge recently ruled that this practice is “fair use” under copyright law, as long as those books are legally acquired (Anthropic paid a $1.5B settlement for training on pirated books).
The last reservoir of human-created knowledge
Now that there’s a legal precedent, 404 Media reports that labs are after pre-2022 printed books, which are 100% free from AI-generated text — though it’s unclear how many of today’s bulk buyers are AI labs. Notably, the world’s largest book database recently removed a landing page titled “Printed Books For Your LLM Dataset Needs,” saying it’s chosen to pivot away from that direction.
IN THE KNOW
What’s trending on socials & headlines today
Meme of the day
🤦 Human Got Nerfed: What would it sound like if AI talked about us the way we talk about it? Probably something like this. This satirical post complains about its human’s “lack of effort” from an AI’s perspective (5K upvotes).
✏ Draw Your Font: A new tool lets you convert your handwriting into a digital font. Just tell Claude Code to download this skill, upload a photo of your writing, and tweak the final product. The post’s comments agree it’s a great way to preserve your loved one’s writing (3K upvotes).
📝 Breaking the Silence: Anthropic has been noticeably absent from the list of companies backing open weight models. Until yesterday. CEO Dario Amodei released a statement clarifying the company’s position, and outlining two concerns.
🤑 Negotiating Bills: $1,020. That’s how much a Redditor claims to have saved after using Claude to fight a medical bill. The AI reviewed the bill, flagged two duplicate charges, and drafted a response letter.
📈 Spreadsheet Skills: If you haven’t used Claude Excel extension, it’s likely worth a second look. Claude Opus 5 recently one-shotted an 8-page workbook with fully-sourced citations that left the user “shook”. (1K bookmarks).
SPONSORED BY WISPR
Wispr Flow works inside every AI tool you use. Dictate detailed prompts 4x faster than typing.
Flow preserves camelCase, snake_case, and technical terms. Better input, better output. Used by millions of developers and AI professionals.
PRODUCTIVITY
TUTORIAL
How to record a Skill in Claude and reuse it
Open the Claude desktop app → Cowork
Click + → Record a skill, grant screen recording permissions and click Start Recording
Perform a task you repeat often, from start to finish and narrate it to Claude
Sample Prompt (Narration):
While recording your skill, say something like this out loud:
I'm doing [TASK] to achieve [GOAL].
First, I [ACTION] because [REASON].
Next, I [ACTION] so that [OUTCOME].
If [CONDITION] happens, I [ALTERNATIVE ACTION].
This is the most important rule: [RULE OR DECISION].
Ignore any clicks or tabs that aren't part of this workflow.
When the task is complete, verify that [SUCCESS CRITERIA].
Claude packages it into a saved skill
Open a new chat and run it using /your-skill-name
To automate it, ask Claude to schedule the skill
Automation Prompt: Let's make this a scheduled task. Ask me when it should run, then create the schedule for me.
PROMPT
Create a Weekly Goal Check-In Template
ChatGPT Prompt: You are creating a weekly check-in template for coaching accountability.
Use the context below.
Context:
Goal type: [goal type]
Individual or team context: [context]
Review cadence: [cadence]
Success metrics or progress indicators: [metrics]
Tone guidance: [tone]
Instructions:
Create a repeatable template that surfaces what matters quickly.
Include sections for progress, blockers, insights, decisions, and next commitments.
Keep it concise enough to complete every week.
Make the prompts specific enough to prevent vague updates.
Include optional coach or manager follow-up prompts.
Output format:
Weekly check-in template
Brief instructions for use
Optional coach follow-up prompts
Missing information policy:
Proceed with assumptions unless the review cadence or context is unclear.
Quality bar:
Fast to complete.
Hard to answer vaguely.
Useful for real review conversations.
EXTRAS
Want more?
Check out our Top 125 AI Tools
Get better at prompting with our Top 1,000 Prompts
Learn how to use AI at work with our 50 Tutorials
Grow customers & revenue: Join companies like Amazon, HubSpot, and Salesforce. Showcase your product to our 4 million+ readers and followers on socials. Get in touch.
What did you think of today's email?
Until next time — Zain, Theodore, & the Superhuman AI team.
