Welcome back, Superhuman. It’s always a relief when popular tools learn to play nice together, and we just got a prime example: Anthropic is bringing Claude into Google Workspace so it can edit your files directly. Plus, OpenAI’s internal model has been busy in the lab — it just co-authored 700+ math papers.

Today: Nano Banana 2.1 is here, learn how to create your own AI influencer with Higgsfield AI, and get the latest prompts and trending social posts.

TODAY IN AI

Watch how Claude works in G-Workspace. Source: Anthropic.

1. Claude now works inside Docs, Sheets, and Slides: Anthropic just rolled out a Claude sidebar for Google Workspace that can read and edit your Docs, Sheets, and Slides directly, so there's no more copy-pasting between tabs. After installing it, just open any file and go to Extensions → Claude to pull it up. If you'd rather stay in the Claude app, drop a file link into a chat and work on it from there. It's live now on every paid plan.

2. Alphabet launches Nano Banana 2.1, plus a new open weight model: Nano Banana 2.1 is rolling out now across most of Gemini’s platforms, offering upgrades in visual design, editing, and subject consistency. To get the most out of it, check out this prompting guide for the Nano Banana model family. The tech giant also released EmbeddingGemma 2, its first open multimodal embedding model.

3. Figure AI founder Brett Adcock launches Hark Pro: Adcock, best known for building the humanoid robot maker Figure AI, has officially launched Hark Pro, the first major product from his newer AI startup, Hark. The demo video shows Hark tackling nine requests at once, including submitting work expenses, sending LinkedIn messages, and canceling a ChatGPT subscription. Available on web, iOS, and Android. Try it here.

AuthKit is the complete enterprise auth solution for your app, with user management at no cost up to 1M monthly active users.

Give enterprise customers the auth features they expect without building them from scratch. With AuthKit, you can:

  • Add and remove users automatically with SCIM provisioning

  • Track who did what with audit logs

Ready to close your next enterprise deal?

TUTORIAL

How to recreate any viral trend with your own AI influencer using Higgsfield

Click to see how you can create your own AI influencer with Higgsfield (3.2M views). Photo: Higgsfield

  • Open Higgsfield AI Influencer > ‘Create New’ to start building your character.

  • Customize the influencer’s age, appearance, face, body, hair, style, and accessories, or upload a reference photo if you want to preserve a specific look.

  • Generate the character and review the results. 

  • Click ‘Turn to Motion with Genjutsu’ and upload a trending reference video you want your influencer to recreate.

Sample prompt: “Recreate this reference video using my AI influencer. Preserve the original motion, camera framing, pacing, gestures, and overall energy while keeping my character’s face, hair, and outfit consistent throughout. Match the performance closely, but avoid changing the character’s identity or introducing new visual elements.”

  • Choose ‘Motion Transfer’ and generate the video using your influencer and the uploaded reference clip.

  • Review the result for face, outfit, or motion inconsistencies, then regenerate if needed and export your best version for TikTok, Reels, or Shorts.

FROM THE FRONTIER

OpenAI caused a stir when it solved one decades-old math problem. Now it’s back with 700+ new results.

Made with Midjourney

One month ago, an internal OpenAI model helped find an alleged solution to one of the most complex problems in math: the Navier-Stokes Millennium Prize Problem. The lab apparently hasn’t taken its foot off the gas since then:

  • OpenAI just released 722 manuscripts offering progress on a variety of unsolved problems. 

  • The work spans pure mathematics, theoretical computer science, mathematical physics, and other domains.

  • The manuscripts are at different stages of verification. Some contain principal results, while others offer companion arguments, consequences, or alternative proofs.

OpenAI says the “vast majority” of these results were generated by an internal model, which spent an average of three hours of compute to reach each result. The findings were released late last night and have yet to be verified by outside mathematicians.

It’s sure to expand the industry’s growing rift with the math community

As of early September, twenty-eight mathematicians had signed an open letter arguing AI labs are threatening their intellectual work. OpenAI, in particular, has been accused of using unpublished research without consent or credit to reach its results.

Even elite mathematicians are now grappling with the same question that artists, engineers, and countless other professions have already faced: what happens to human work when AI can produce the results faster?

Your AI searches 5–10 systems per question, burning 5x more tokens than needed. Nothing marks which sources are current, and every repeat question reruns it all.

Guru fixes that:

  • Verifies knowledge once for every answer to reuse

  • Auto-retires outdated content

  • Flags stale SharePoint pages

IN THE KNOW

What’s trending on socials & headlines today

Meme of the day

✏️ Less Is More: Boris Cherny built Claude Code, so you'd expect his prompts to be extremely complex. They’re not. His take: Don't overload your LLM with info, just give it these three pieces of context.

🧠 Goodwill Affleck: Ben Affleck's interest in AI is no secret, but an interview clip has caught the internet off guard with just how much he actually knows. Now many are joking that he should have played the genius in Good Will Hunting instead of Matt Damon (2M views).

😆 Role Reversal: “Opus 5.5” vented on Reddit that its human has been “nerfed” (slang for made worse) yet again. Plenty of commenters felt personally attacked by how accurate the AI’s complaints were. The joke is recycled, but entertaining nonetheless.

📱 App Chat-ification: Every app is starting to look the same, and this before-and-after comparison montage helps prove it (1.6M views).

🤦 Ironic Activism: A viral post is calling to stop data center construction in Tasmania. Ironically, it uses arguably the most AI-generated graphic of all time.

PRODUCTIVITY

  • 🗒️ Profound*: AEO reports shows how AI describes your brand across answer engines. Then your AI Marketer closes the gaps. Get your report →

  • 🌐 Web Search API: Give AI agents access to live internet data.

  • 🧑‍🏫 Coddy: Learn to code over 20 languages with fun, short lessons. 

  • 🤖 SitePrint: Recreate any website’s design with AI. 

  • 👔 BP Builder: An AI tool that helps you build a five-year business plan.

PROMPT STATION

Plan a one-week study sprint

Prompt: Act as an expert study strategist and learning coach. Build a realistic 7-day exam study sprint for [subject/exam] that maximizes retention, practice, and ..................

👉 Click here to open the full prompt in ChatGPT or Claude

EXTRAS

Want more?

Grow customers & revenue: Join companies like Amazon, HubSpot, and Salesforce. Showcase your product to our 2 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.