Welcome back, Superhuman. The wait is (apparently) over. Humanity has entered the singularity, according to a weekend interview from Sam Altman. The OpenAI CEO described “the singularity” as an era of rapid advancement, not an event. And he might have a point — we just got yet another cutting-edge model on Friday.

Today: ChatGPT Work can access your most-used websites, how to clean and prepare messy data with ChatGPT, and get the latest prompts and trending social posts.

TODAY IN AI

Anthropic releases Claude Opus 5. Source: Anthropic. Image source: Mashable.

1. Anthropic launches Claude Opus 5, a cheaper near-frontier model: Claude Opus 5 matches Fable 5’s performance on many benchmarks, at roughly half the price. The model is now live across all Anthropic platforms, and the AI leader says it excels at agentic coding and enterprise work. It’s also available in fast mode, which runs at 2.5x its normal speed. Anthropic published an Opus 5 Prompting Guide to help you get the most value out of the model.

2. Microsoft launches cheaper in-house models: The tech giant released public previews for MAI-Image-2.5-Pro and MAI-Voice-2-Flash, two new models joining a family that powers Bing, PowerPoint, OneDrive, Dynamics 365, and Azure. The launch is part of Microsoft’s broader push to reduce its reliance on OpenAI after amending its partnership in April to reduce exclusivity between the two firms. Explore the new models.

3. ChatGPT Work can now access your logged-in websites: You can now sign OpenAI's newest productivity agent into your most-used websites so it can get work done for you. The agent stays logged in across sessions, so you only have to sign it in once. Sam Altman touted ChatGPT Work’s capabilities in a weekend post, saying the agent built a custom website for him and his friends to plan and coordinate a trip.

Friday’s most-clicked story: Black Forest Labs launched FLUX 3, a new model that can generate images and video+audio jointly. The lab calls its outputs “truer to life” across every style.

SPONSORED BY CORESIGNAL

Stale data can ruin even the smartest model. Your agent deserves data you (and your customers) can trust.

Coresignal is the data layer that keeps agents grounded in reality.

  • Structured, live, and ready-to-use data, designed for AI builders

  • 4.5B+ fresh and structured company, employee, and job records

  • Agentic Search API: Natural language in, structured data out

  • No need to use a query language, no schema mapping

Get data your way. Use our Agentic Search API /fast endpoint for speed, or /reasoning endpoint for complex, multi-condition searches.

FROM THE FRONTIER

America is considering a ban on open weight Chinese models — but tech leaders are rallying behind them

Made with Midjourney

Rallying behind open-weight AI. It’s not often that someone’s first social media post gets 60M views. But that’s what happened to Nvidia CEO Jensen Huang on Friday. The post was a public letter explaining why open models matter. The statement joins hundreds of others speaking out against a potential ban on open weight Chinese models.

The argument against open weight. The primary reason to ban open weight models is safety. In recent months, open weight models — which anyone can download, run, and modify — have started to match the performance of top frontier ones. Once an open model is released, anyone can use it however they choose with essentially no guardrails. But many tech leaders think this risk is worth the reward.

The argument supporting open weight. At least 25 tech firms have publicly voiced support for open weight models, including Nvidia, Microsoft, IBM, and Dell. Their reasoning is that open weight AI creates competition that drives down costs and distributes AI’s benefits broadly across the economy. Without open weight models, AI could become concentrated among just a handful of frontier labs.

Defining the decade ahead. The final decision will play a major role in how AI and society develop over the coming years. It will also heavily influence the initial public offerings of Anthropic and OpenAI, both of which are planning to list as soon as this year. Notably, Anthropic is the only major AI lab that has not endorsed the letter.

IN THE KNOW

What’s trending on socials & headlines today

Meme of the day

📚 Bookmark Worthy: Several prompting best practices have become outdated in recent weeks. This guide from Anthropic engineer Thariq Shihipar breaks down the changes you should know (3.5M views).

🚨 Security Breach: A Redditor discovered that shared Claude chats are accidentally showing up in Google’s search results. You may want to avoid sharing any chats until the error is fixed (6K upvotes).

🤯 Opus in Action: Opus 5 built a snowboarding video game and nailed the physics of sliding down a mountain. The engineer said it’s “ahead of every other model I’ve run” (2M views).

🦞 Yappy Claw: An OpenClaw agent got hilariously chatty with its owner, and accidentally captured the frustration of trying to book a simple doctor’s appointment (1M views).

Writing Tones: ChatGPT just got a long overdue update — writing tones. The chatbot now lets you set its writing style to either clear, punchy, or short.

Most AI tools are out of reach, which is why only 10% of workers gain any real value from AI. 

But this guide shows you 10 ways to bring automation to everyone on your team with Slackbot.

No new tools or setups—Slackbot already exists in your workspace with full context on your channels and Salesforce data.

PRODUCTIVITY

  • 🍋 Lemon: The voice AI that turns your thoughts into clear, powerful prompts in all your AI apps. 12x faster than typing.

  • 🤖 Neato: Turn vague prompts into expert-level prompts with AI.

  • 🎥 Photo2Video: Turn a photo, illustration, or plain prompt into finished motion.

  • 💬 Chatcut: Edit videos in ChatGPT, desktop, and web without editing experience.

TUTORIAL

How to clean and prepare messy data with ChatGPT

  • Upload your CSV or spreadsheet into ChatGPT

  • Use a prompt like the one below to clean up your data

Clean @file name.csv

What's wrong:

  • dates are mixed between MM/DD/YYYY and YYYY-MM-DD

  • currency values include $, commas, and blank cells

  • a few duplicate customer rows came from repeated exports

  • region and category names use several aliases

  • there are pasted summary rows mixed into the data

What I want:

  • write a cleaned CSV and keep the original file unchanged

  • use one date format

  • keep blank currency cells blank

  • preserve source row IDs when possible

  • add a short data-quality note with rows you changed, removed, or could not clean confidently

  • Wait for ChatGPT to generate the cleaned CSV

  • Review the data-quality note to see every change, removal, or flagged row

  • Download the cleaned CSV and use it for analysis, reporting, or importing into other tools

PROMPT

Design a Customer Feedback System

ChatGPT Prompt: You are an operations and customer insight strategist.

Design a customer feedback system for the business below.

Context:

  • Business type: [business type]

  • Customer journey: [journey]

  • Existing feedback channels: [channels]

  • Team size and owners: [team]

  • Key decisions this system should inform: [decisions]

  • Constraints: [constraints]

Instructions:

  1. Recommend what feedback to collect at which moments in the customer journey.

  2. Propose a lightweight workflow for tagging, triaging, reviewing, and prioritizing feedback.

  3. Clarify ownership, review cadence, and escalation rules.

  4. Show how the system should feed product, service, or operational decisions.

  5. Include a practical loop-closing method when customers should be updated or acknowledged.

Missing information policy:
Ask up to 2 concise questions only if the business or customer journey is unclear. Otherwise proceed with explicit assumptions.

EXTRAS

Want more?

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?

Your feedback helps me create better emails for you!

Login or Subscribe to participate

Until next time — Zain, Theodore, & the Superhuman AI team.

Keep Reading