|
|
|
|
Welcome, humans. |
Most courses get built behind closed doors, then dropped on students who had no say. We tried the opposite with The Neuron Academy. |
If you haven't met it yet, The Neuron Academy is our AI training site, built by the team behind this newsletter and The Neuron: AI Explained podcast. It's a library of short, beginner-friendly courses on the stuff you actually do at work: writing better prompts, getting more out of ChatGPT, Claude, and Gemini, researching faster, and automating repetitive tasks. Each lesson takes about 15 minutes, and you don't need a technical background. |
We opened it early to a select group of Neuron subscribers and put them to work. Tyler was one of them, and the feedback he gave us along the way helped shape what the Academy is today. (Turns out the best curriculum committee is the people taking the classes.) |
His review of the finished product: |
|
|
Readers like you, building something for readers like you. See the result with our new course on building branded scientific visuals with Claude and Claude Cowork. |
Here’s what happened in AI today: |
😺 OpenAI says its AI disproved a famous computer science hypothesis 📰 Anthropic filed for an IPO with an $80.75B net loss 📰 Microsoft offered voluntary exits to about 60,000 workers 🍪 Ex-GitHub CEO launched Entire, a Git forge for coding agents 🎓 Race two AI models on the same boring task
|
... and a whole lot more that you can read about here. |
|
|
|
|
😸 OpenAI Says Its AI Disproved a Famous Computer Science Hypothesis, Then Launched GPT 6.1 Terra |
On Monday, OpenAI put out two announcements that look like they came from different companies. One reports three new discoveries in math, computer science, and physics. The other introduces GPT-6.1 Terra, a cheaper model built to click through office software. |
Here's what happened: |
OpenAI says its models proved the local smoothing conjecture, a 1991 math problem about how waves focus, in every dimension. It says its system disproved the Exponential Time Hypothesis, and a proof-checking program called Lean verified the result. It says its models produced a rigorous existence proof for quantum chromodynamics, the theory of the strong force. Terra costs $1 per million input tokens and $5 per million output tokens, half the price of GPT-6.1 Sol.
|
Start with the computer science result, because it comes with the weirdest cast of characters. The Exponential Time Hypothesis (ETH) is a belief about a puzzle called Boolean satisfiability: given a pile of true/false switches and a list of rules, can you find a setting that satisfies every rule? The best known methods get exponentially slower as you add switches, and the ETH says nothing fundamentally faster exists. |
Hundreds of papers use it as a starting assumption, in fields from databases to computational biology. If the ETH is false, a lot of "can't be done faster" results become open questions again. |
OpenAI says its system found the disproof on October 16, and Lean checked it the next day. Lean works like a spellchecker for logic: it confirms every step follows from the last. A working group of experts now has to confirm the claim means what we think it means. It's led by Russell Impagliazzo, Ramamohan Paturi, and Ryan Williams, and Impagliazzo and Paturi are two of the three people who originally proposed the ETH. So the authors of the hypothesis are now deciding whether a robot just knocked it down. That's a fun group chat. |
OpenAI is upfront about how raw this is. It calls the manuscripts raw research data, and it says turning a disproved ETH into faster software will take significant further research. A group led by Nobel laureate David Gross is also holding back the physics result while it writes its own account. |
Same day, very different launch: GPT-6.1 Terra. Terra is the middle model in OpenAI's lineup, built for computer use, meaning it looks at a screen, then clicks, types, and moves between apps like a person would. OpenAI's own example is reading incoming invoices and entering the details into accounting software. Riveting? No. Valuable? Ask anyone who has keyed invoice numbers by hand on a Friday afternoon. |
OpenAI claims Terra scores 15 points higher than its predecessor on its Agents' Last Exam at about a third of the cost. It says Terra matches Astra at one-tenth the price per token and finishes over three times faster than Astra and Sol. An independent NerdsChalk test saw a smaller gap, with Sol taking about 1.8 times as long on three short tasks. An OpenAI engineer called it more of a way to stretch usage limits than a bombshell. |
Both launches share one problem: how do you check the work? A filled-out form takes seconds to verify. A proof of a 35-year-old conjecture takes months, a working group, and a conference OpenAI has agreed to fund. Watch for that group's verdict on the ETH, since it's the cleanest test yet of whether AI-generated proofs hold up when the people who built the field take a look. |
|
|
FROM OUR PARTNERS |
Meeting notes from your wrist |
|
|
Not every meeting has a calendar link. Granola, the AI notepad for people with back-to-back meetings, is now on Apple Watch. Tap to start a note on the way into a coffee, a site visit or a hallway chat. Granola transcribes the conversation and turns what you jot down into summaries and next steps, waiting on your laptop when you're back. |
Download Granola and try it in your next meeting.
3 months on us with the code THENEURON. |
Try Granola |
|
|
🎓 AI Skill of the Day: Race Two Models on the Same Boring Task |
OpenAI's Victor Nunez raced Terra against Sol to show off Terra's computer use. He had each model fill out the same long web form in Fast mode while the screen recorded, and according to Gigazine, Terra finished more than two minutes sooner. |
NerdsChalk ran a smaller version in Codex: a 15-field form, three runs per model. Terra and Sol both got every field right every time, and Terra's median time was 9.4 seconds to Sol's 17.3. The cheapest model, Luna, got every field right in only one of three runs. |
That result is the lesson. OpenAI says Terra is over three times faster on its benchmark, but NerdsChalk's gap was closer to 1.8 times, and the model with the lowest price slipped on accuracy. Vendor numbers are averages across their tasks, and your task is the one that matters. You can run the same race on any chore you'd hand to AI: |
Pick a task you can grade, like filling a form with fake data, cleaning ten spreadsheet rows, or logging five sample invoices. Use a throwaway form and dummy data, since agents that click around the web can be tricked by what's on a page. Run identical instructions three times on a cheaper model and three times on a pricier one. In ChatGPT Work or Codex, Terra shows up in the model picker for Plus, Pro, Enterprise, and Edu plans. Score two things: how many runs came out fully right, and how long each took. Switch to the cheaper model only if accuracy holds.
|
Fill out this form using the fake data below. Do not submit it. When you finish, list any field you were unsure about and tell me how many steps it took.
|
Think of it as a taste test, except this one has a stopwatch. Three runs won't settle anything scientific, but a 9-second versus 17-second gap on your own chore tells you what the benchmark chart can't, and one wrong field in three runs tells you the cheap model isn't ready for that job yet. |
Have a specific skill you want to learn? Request it here. |
|
|
FROM OUR PARTNERS |
Meta's Muse Agent Is Visiting Your Site. Can You Detect It? |
|
|
Meta's Muse agent browses websites, logs into accounts, and completes purchases. Fingerprint identifies Muse traffic, provides visibility into its activity, and lets you control it with custom policies. |
Free detection for existing customers. |
Learn How to Detect Muse |
|
|
🍪 Treats to Try |
*Asterisk = from our partners (only the first one!). Advertise to 700K+ readers here! |
*What Is Data Management? A Guide to Systems, Processes, and Tools. Learn more Entire is a code-hosting site from GitHub's former CEO where review agents score every push you make, and it can mirror your existing GitHub repos. Oracle's Fusion apps now include 25 task-doing agents for finance, HR, supply chain, and customer experience, with access rules enforced inside the database itself. Strawberry is a browser where shareable agents called Companions research sales leads in parallel, transcribe meetings, and run recurring web chores. SurgePix generates visuals where text and regions stay editable afterward, so you can fix one headline instead of re-rolling the whole image. MiniMax M3.1 Flash is a low-cost model for agents and coding that MiniMax says outputs more than 200 tokens per second. ManyAgents is a Mac app that splits coding work across Claude Code and Codex agents in separate working copies of your repo, then adds review and browser testing. Hob is an open-source desktop assistant that picks and runs an open model on your own computer, so you can work through PDFs offline.
|
|
|
📰 Around the Horn |
 |
Somewhere, an engineer who spent four years learning CAD software just felt a chill. |
|
Anthropic filed to go public as ANT, reporting $16.29B in first-half revenue and an $80.75B net loss, including a one-time $72.4B accounting charge (AP). Meta launched about $35B in stock offerings as part of a $95B AI funding plan, with Nvidia, AMD, and Broadcom signaling interest in up to $5B of shares. Microsoft offered voluntary exits to about 60,000 eligible workers, roughly 27% of staff, per Reuters and AP; how many accept is up to them. SpaceX signed a roughly $40B GPU lease led by Apollo, the largest GPU financing so far, with deliveries starting in the first half of 2027. The FTC demanded testimony from Anthropic executives in its AI safety probe, which already includes document demands to Anthropic and OpenAI, Axios reported. Cloudflare disclosed a data leak in which 155 responses from its hosted GLM-5.3 model included other customers' recent data between Aug. 27 and Sept. 9; it has since patched the bug.
|
Want absolutely EVERYTHING that happened in AI this week? Click here! (ATH digest URL needed) |
|
|
UPCOMING EVENT: The Neuron IRL in San Francisco |
|
|
The Neuron is going IRL on November 18 at 4:30 PM PT! Join us in San Francisco for drinks, bites, special guests, and a live recording of The Neuron podcast. Huge thanks to Slack and Alumni Ventures for making it happen. |
It’s free, but space is limited. Save your spot! |
|
|
|
|
|
|
Anthropic just filed to go public, and the paperwork doubles as a forecast: within a few years, AI could match top human experts. |
Here’s what they’re predicting: |
AI that matches or beats the best experts in most professional and scientific fields. AI that works alone for hours, days, or weeks, then checks in for feedback. AI that runs lab equipment and robots through digital controls.
|
Anthropic’s name for the endgame is “a country of geniuses in a data center,” meaning millions of AI agents working at once on software, medicine, physics, and more. |
Why take it seriously? Anthropic says it’s already living it. Over 90% of September’s code contributions were written by Claude, and each engineer ships over 45x the daily code they did in 2024. Revenue is up about 1,200% in a year while headcount rose about 170%. |
It’s a flywheel: better models help build the next, better model. Anthropic calls it “closing the loop.” |
It also expects demand to outrun computing power for a while, since the economy has barely tried what today’s models can do. |
Big caveat: this is a company’s own forecast, in a document meant to attract investors. Read it like a pitch deck, not a weather report. |
|
|
New from The Neuron: AI Explained |
 |
Click the image to watch the video. |
|
New episodes air every week on Wednesdays: Spotify | Apple Podcasts | YouTube |
|
|
|
|
|
|
|
|
That’s all for now. If you want to get featured above, fill out the poll below and tell us how we did today! |
|
What'd you think of today's email? |
|
|
|
|
Btw: We just launched a robotics newsletter! Sign up for it here. |
P.S. Love the newsletter, but only want to get it once per week? Don’t unsubscribe—update your preferences here. |
No comments:
Post a Comment
Keep a civil tongue.