Welcome back, Superhuman. Just last week, Sam Altman declared that humanity has entered the singularity. He might’ve actually been right. An unreleased OpenAI model just solved 10 famous open research problems — a run that arguably surpasses any single human mathematician in history.
Today: Gemini Spark can browse the web, how to design high quality websites with Claude Skills, and get the latest prompts and trending social posts.
TODAY IN AI
1. Unreleased OpenAI model pushes the boundaries of human knowledge: OpenAI’s internal Astra model solved 10 open problems spanning mathematics, quantum complexity, and theoretical computer science, all supported in a 249-page paper. Total estimated token cost was $2,000. Prior to this, none of the problems had seen new progress on their main results in at least a decade. Sam Altman was reportedly in Washington, DC last week demoing Astra to federal officials.
2. Google’s agentic assistant can now browse the web using Chrome: With your permission, Gemini Spark can complete online tasks using your logged-in accounts and saved passwords to do things like research flights or hunt for new apartments. See how auto browse works. In a separate update, Google now lets you summon Gemini at any time by pressing the FN key in the macOS Gemini app.
3. AI startup raises $200M to simulate Earth’s eight billion people: After launching five months ago, Simile has closed a Series B round with the goal of accurately predicting human decisions at scale. The startup, built by Stanford researchers on their own academic work, has released two models: a foundation model trained on real people, and a confidence model that estimates accuracy. See how it works here.
Friday’s most-clicked link: This viral prompt helps your chatbot write more like a human. Just upload it before every conversation or permanently add it to your LLM’s instruction file.
SPONSORED BY AWS
AI agents need their own identity — not service accounts. This workshop covers agent identity and access management on AWS, plus technical demos of Auth0 and CyberArk PAM.
You'll learn how to:
Scope time-bounded tokens that expire per action with AWS Security Token Service
Configure automatic rotation using AWS Secrets Manager
Implement AWS CloudTrail to trace each API call to identity, session, and task
FROM THE FRONTIER
The busiest week of earnings season just ended — here’s what it told us about the state of AI and the economy

Made with Midjourney
Big Tech’s earnings in five words? Dwindling cash, soaring compute costs
We’re witnessing history. Amazon and Alphabet — two of the most profitable companies ever — reported negative free cash flow last quarter, citing the hundreds of billions being poured into AI infrastructure (this chart puts it in perspective). Meta wasn’t far behind, with cash flow plummeting 91%. But fortunately, the declining cash flow didn’t spook Wall Street.
But revenue is strong and (mostly) supporting AI’s growth
The entire Magnificent 7 posted double-digit annual revenue increases, except Nvidia who reports in a few weeks. The strong results sent the tech-heavy Nasdaq up roughly 2.5% on the week, reversing a downward trend that started in June. Notably, Microsoft soared 19% and added a single-day record of $450B in market cap, while Meta and Apple declined substantially. The biggest recurring talking point? RAMageddon.
The global memory crunch is hitting Big Tech harder than ever
The ongoing memory shortage, nicknamed RAMageddon, is still driving up costs for both Big Tech and consumers. Tesla CEO Elon Musk called current memory pricing “pretty insane,” while Apple CEO Tim Cook referred to it as a “hundred-year flood.” It’s already forced price increases for Macs and iPads, and analysts expect more on the way.
IN THE KNOW
What’s trending on socials & headlines today

Meme of the day
👤 TIME Headshots: Is your LinkedIn photo due for an update? This viral prompt transforms regular selfies into editorial-style headshots you’d actually post, based on Marco Grob’s TIME cover portraits (2K upvotes).
👀 Sam’s Suggestion: Sam Altman accidentally drew the internet’s ire on Friday when he posted a “cool use case” for ChatGPT. Many people did not think it was cool and flooded the post with sarcastic comments (12M views).
🚨 Be Aware: Don’t believe any satellite images you saw over the weekend. Google Earth got a new AI image-generation feature that was abused so quickly the company had to revoke it within a day. Read about the short-lived update here.
📝 Leopold’s Letter: Leopold Aschenbrenner’s AI fund just had its roughest month since launching. He sent out an investor update that’s being applauded for its calmness under pressure, along with a mild humblebrag about the fund’s YTD performance (3M views).
SPONSORED BY CORESIGNAL
Stale data can ruin even the smartest model. Coresignal is the data layer that keeps agents grounded in reality.
Designed for AI builders
4.5B+ fresh company, employee, and job records
Natural language in, structured data out
Use our Agentic Search API /fast endpoint for speed, or /reasoning endpoint for complex, multi-condition searches.
PRODUCTIVITY

TUTORIAL
How to design high quality websites with Claude Skills

Go to Refero Design, browse designs and pick a website design you like
Download the Design.md file for the design system you want to use
Open the Claude Desktop app
Go to Settings → Customize → Skills → Add Skill
Upload the Design.md file and enable the skill by toggling it ON
Switch to Cowork mode in the Claude app
Add ‘/‘ to the prompt and select the skill from the dropdown menu to activate that design system
❝Sample Prompt: Build a personal portfolio website using the /[enter skill name]
PROMPT
Draft a constructive feedback conversation
ChatGPT Prompt: You are drafting a constructive feedback conversation.
Use the context below.
Context:
Conversation context: [context]
Behavior or issue to address: [behavior]
Impact of the issue: [impact]
Desired change: [change]
Relationship tone to preserve: [tone]
Stakes: [stakes]
Instructions:
Make the feedback specific and behavior-based.
Avoid vague praise-sandwich filler.
Include language that opens dialogue without weakening the message.
End with a concrete next-step discussion.
Keep the tone firm and respectful.
Output format:
Opening
Main feedback message
Clarifying questions
Next-step discussion
Missing information policy:
Ask first only if the behavior or desired change is unclear.
Quality bar:
Specific and respectful.
No evasive wording.
Clear next steps.
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.





