Did you know that we have LinkedIn, Instagram and X accounts
that you can follow?
In this issue:
Hi {{first_name|everyone}},
OpenAI says an internal model came up with a broad set of new math results, and what caught our eye is how much effort they're putting into letting people actually check the work.
We also have a quick word from our friends on how FDEs are bridging the AI deployment gap plus an idea for juggling multiple AI models, a few handy tools, and an artwork a whole museum can control at once. Alright, let's get right into it.
On to the good stuff.
🤿 DEEP DIVE
AI math, with receipts
The results are interesting. How they’re being released might be too.
OpenAI says one of its internal frontier models has produced a broad range of new mathematical results. But what caught my attention is how much effort is going into making those results inspectable.
They’re being published in a GitHub repository, alongside formalizations of many proofs in Lean, which allows the proofs to be checked by a computer.
And OpenAI isn’t only publishing the answers.
The repository includes 10 summaries of the model’s reasoning, statistics on how many problems were attempted, and estimates of how much compute went into the results. On average, each result used roughly the equivalent of 3 hours of ChatGPT Pro thinking.
That’s a pretty useful detail. Instead of just seeing what the model eventually produced, the math community gets at least some visibility into the work behind it.
There’s also a bigger process being built around this. OpenAI says it consulted the independent Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study, is using protocols for revisions and citations, and is exploring other community-hosted ways to release the work.
So this isn’t simply “AI solved some math.”
It’s also an experiment in what happens after an AI produces a potentially important scientific result. How do you publish it, check it, explain it, revise it, and give researchers enough information to actually understand where it came from?
OpenAI says future releases will keep improving that process, including the citations, mathematical exposition, and presentation.
The quick take
An internal OpenAI model produced a broad range of new mathematical results.
Many proofs are being formalized in Lean for computer checking.
• OpenAI is also releasing reasoning summaries, attempt statistics, and compute estimates.The average result used roughly three hours of ChatGPT Pro thinking worth of compute.
OpenAI says it plans to keep improving how major AI-produced scientific results are disclosed.
🤝 FROM OUR FRIENDS
AI’s Next Bottleneck Is Deployment.
New models are easy to demo. Making them work inside real customer operations is not.
That is where forward deployed engineers come in: bridging code, customer context, and production outcomes.
The free State of FDE Jobs 2026 report maps the emerging labor market around this work.
🔥 WHAT’S THE BUZZ
What people are building, sharing, and talking about
As people accumulate subscriptions to several AI tools, choosing the right model becomes work of its own. This suggests a different setup: instead of you deciding “Claude or Codex?” every time, a cheap decision layer could make that choice automatically based on the task.
⚒ TOOL SNAPSHOTS
Futuristic tools within AI, no-code, and productivity
Give agent runs a shared record of who did what.
Why it’s useful: I can see this being handy when one agent run crosses several systems and you need a consistent way to trace who initiated it and what it cost.
🧩 Fuse AI
Run more of your GTM workflow from one stack.
Why it’s useful: Useful for developers and agents that need enrichment, outreach, automations, and workflows without stitching together a separate product and API for each step.
Turn messy business documents into structured data through one API.
Why it’s useful:Keeps document processing simpler by handling parsing, validation, edge cases, and feedback through a single endpoint.
🤖 AI MADE THIS
Interesting and inspiring creations made with AI
Image credit: MMoCA
Diffuse Control - an AI artwork that an entire museum audience can manipulate
Beeple's Diffuse Control isn't an AI image hanging on a wall. It's a 12-screen artwork that keeps changing as museum visitors manipulate it together, with Stable Diffusion remixing works from the museum's collection in real time. The new installation opened in Madison last month, turning the audience itself into part of the generative process.
AI used
Stable Diffusion is confirmed as the primary image-generating model inside Beeple's custom software environment. An OpenAI system is also used for a limited title-generation function.
ℹ️ ABOUT US
The Intelligent Worker helps you to be more productive at work with AI, automation, no-code, and other technologies.
We like real, practical, and tangible use-cases and hate hand-wavy, theoretical, and abstract concepts that don’t drive real-world outcomes.
Our mission is to empower individuals, boost their productivity, and future-proof their careers.
We read all your comments - please provide your feedback!
Did you like today's email?
What more do you want to see in this newsletter?
