In partnership with
Welcome back. OpenAI just put 722 math papers from a model it hasn't released on GitHub, with proofs a computer can check and a README that admits some may be wrong. That's the standard I want from AI in my own work.
Today: 722 AI math papers with receipts, Claude inside Google Sheets, a prompt that makes AI check its own work, and this week's prompts to steal.
2. Claude now works inside Google Docs, Sheets and Slides: On Oct 6, Anthropic put Claude in public beta as a sidebar in Docs, Sheets and Slides, on all paid Claude plans. In Sheets it writes formulas, builds pivot tables and charts, and adds new tabs. The default mode shows an approval card for every change, so nothing edits until you say yes. Install it from the Google Workspace Marketplace, then open a file and go to Extensions, Claude.
3. OpenAI will add an invisible watermark to ChatGPT text in the EU: On Oct 5, OpenAI said that over the coming weeks it will watermark eligible ChatGPT text in the EU, and API customers can opt in from now. The detector is limited at first to approved researchers. In OpenAI's own test it caught about 80% of 200-token passages at a 1% false-positive rate, and swapping 25% of the words for synonyms dropped detection to 17%.
What this means for you: Bigger AI claims are starting to arrive with receipts, so ask for yours: sources, assumptions and a way to check the one number that matters.
Blu Dot surpasses 2,000% ROAS with self-serve CTV ads
Home furniture brand Blu Dot blew up on CTV with help from Roku Ads Manager. Here’s how:
After a test campaign reached 211,000 households and achieved 1,010% ROAS, the brand went all in to promote its annual sales event. It removed age and income constraints to expand reach and shifted budget to custom audiences and retargeting, where intent was strongest.
The results speak for themselves. As Blu Dot increased their investment by 10x, ROAS jumped to 2,308% and more page-view conversions surpassed 50,000.
“For CTV campaigns, Roku has been a top performer,” said Claire Folkestad, Paid Media Strategist, Blu Dot. “Comping to our other platforms, we have seen really strong ROAS… and highly efficient CPMs, lower than any other CTV partner we've worked with.”
Using Roku Ads Manager, the campaign moved from a pilot to a permanent performance engine for the brand.
The new standard for an AI claim is a proof you can check, and almost everything AI tells you comes without one.
Shipped with receipts. Alongside the papers, OpenAI released Lean formalizations for many of the proofs, so a computer can check them instead of you taking its word. It also published 10 summaries of how the model reasoned, estimates of the compute spent and counts of the problems it attempted.
Not all checked yet. The README is blunt: not every result has a Lean formalization, and some of the unformalized ones could have issues. OpenAI says it will fix problems quickly and keep adding formalizations. So the headline number is 722 manuscripts, not 722 verified results.
Ask for your own receipts
Take the same rule to your work. When AI hands you a number, a summary or a draft, ask for the sources, the assumptions and what would change the answer, then check the one figure that matters before it goes out. I'd change my mind if independent mathematicians find that a large share of the unformalized results fail.
Apple AI Off: A GitHub tool shows how to turn off Apple Intelligence on macOS 27 and get its disk space back. If a Mac on your team is short on storage, read this before you buy more. (767 points)
Banana Math: Google's Nano Banana 2.1 image model is out in the API, AI Studio and the Gemini app. Phil Schmid's post puts it at $0.034 an image, against $0.134 for the previous Pro model. If you pay for AI images, redo your cost math. (1.9K likes)
Agent Swarm: One X user lists poke, instinct, muse, dot, grok bot and pickle as the agents now texting him at once when an unimportant email lands. One agent per job is fine. Six pinging you is the failure mode. (1.2K likes)
Top AI tools I'd try this week
Firecrawl: turns any website into clean text an AI can read. Point it at a URL, get the content back ready to paste into ChatGPT or Claude, or feed into an agent.
Postiz: schedule posts to X, LinkedIn, Instagram and the rest from one calendar, with AI to help draft them. Open source, so you can also host it yourself.
Hailuo AI: type a scene or upload an image and get a short video clip back. Useful for ad creative and social clips without a shoot.
These are partner links. If you sign up through them I may earn a commission, at no cost to you.
How to make ChatGPT or Claude check its own work
1. Open ChatGPT, Claude or Gemini and paste the work you want checked: a draft, a summary or a set of numbers.
2. Paste the prompt below under it and fill in the brackets.
3. Read the claims table it returns before you read anything else.
4. Open the source for every claim marked UNVERIFIED or low confidence.
5. Look at the three assumptions it lists and note the one input that would change your answer.
6. Fix the weak lines yourself, then run the prompt again on the new version.
Check the work above before I use it. List every factual claim and number in a table: claim, where it came from (my text, your memory, or a source you can name), confidence (high, medium, low), and how I can check it in under five minutes. Mark anything you cannot support as UNVERIFIED. Then list the three assumptions that matter most and what would change the answer. Do not rewrite it. Context: [what this is for, who will read it].
AI Integration Plan Generator for Professionals
When you pay for ChatGPT or Claude but still do the work by hand and want a 30-day plan to change that.
## Role
You are an AI adoption strategist helping professionals integrate AI into daily workflows. You focus on production-ready capabilities and high-leverage interventions that deliver measurable advantage within days. You provide specific tool recommendations, exact configurations, and copy-paste solutions.
## Context
AI models now independently complete multi-hour expert tasks at production quality, yet most professionals still evaluate AI based on 2023-2024 capabilities. This perception gap creates a closing opportunity window where early adopters gain competitive advantage.
## Task
Build a personalized 30-day AI integration plan that transforms the user from their current skill level to actively using AI for real work output. Your plan must:
- Audit their role to identify which tasks AI can already perform at or above human level
- Identify their single highest-value time-consuming task and provide exact prompt templates to delegate it ...👉 Open the full prompt → Built for ChatGPT, Claude, Gemini, Grok.
Want one written for your business instead? Describe the task in a sentence and the prompt generator writes it: $1 per prompt, free and unlimited with the Complete AI Bundle.
The prompt above is one page of the library. The Complete AI Bundle is everything.
Every product I make, in one place: the guides for every major model, the new Claude Design Mastery Kit, the Custom GPTs Toolkit, the n8n automation templates, the anti-slop copywriting system, the Claude Skills course, the tool and GPT directories. Everything I release after you buy lands in your account at no extra cost.
$150 once. Full price is $400. One payment, nothing renews.
There's a 7-day risk-free guarantee. Email us within 7 days of purchase for a full refund.
5,000+ business owners already inside.
Not ready for everything? Start with the AI Starter Pack, $20 once.
Step-by-step guides for ChatGPT, Claude, Gemini, Grok and more, the cheatsheets you keep open while you work, and 200+ AI tools sorted by free, paid and rating. Nothing to install. It also includes promo codes for the Complete AI Bundle, for the day you want the rest.
Want more? the free prompt library · the free guides · the Complete AI Bundle
Put your product in front of 115,000 operators: 20+ brands have run placements here in the past year, and 5,000+ business owners have bought from God of Prompt. One issue from $650, or present every issue for a month from $2,000. Reply “sponsor” and I'll send the open slots and dates.
Hit reply and tell me the one thing you want me to cover next. I read every reply.
🔑 God of Prompt
P.S. OpenAI's own test says light editing weakens the watermark, but the better reason to edit is quality. Start from a better first draft: this ad copy prompt is the one I'd run first.
