In partnership with
A Single Typo Shouldn’t Cost You a Sale
A shopper types “running shos” and gets nothing back. Another types “red waterprof trail runners under $150” and hits the same wall. Both knew what they wanted.
Traditional site search relies on exact keyword matching, leaving shoppers stranded at “0 results found” screens.
Gladly understands what shoppers mean, typos and all. Instead of a dead end, they get products that fit what they asked for and a conversation that keeps them shopping.
Beginners in AI
Good morning and thank you for joining us again!
A new test dropped AI into real expert work, the kind bankers, lawyers and chip designers do, and the best model finished about 1 in 8 tasks. That's a handy number to remember the next time a chatbot sounds sure of itself. Today's edition shows where AI already pulls its weight and where you'll still want to check its work. As always, every story here is human picked and edited.
THE FRONT PAGE
The Best AI Finished 12% of Real Expert Work in a New Test
TLDR: A new test called GDPx gave top AI models real expert work in 11 fields, and the best one, Anthropic's Claude Opus 5.5, finished just 12% of it.
The Story:
SemiAnalysis, a research firm, published GDPx on November 2. It drops AI agents into realistic work, like a deal room of about 2,000 files from a real private equity deal or real lab data for biology, with up to 8 hours per task across 11 fields from chip design to law.
Claude Opus 5.5 topped the ranking at 12%, with Google's Gemini 4 Carbon best at biology and OpenAI's GPT-6 Astra best at materials science. Finance and legal were so hard that nearly every model scored exactly zero under pass-or-fail grading, so the authors added partial credit there, where Opus averaged 44% on legal and Astra 25%.
Its Significance:
A 12% score doesn't make AI useless at work. It means today's best models are quick first-draft helpers, not someone you hand a whole job to and walk away from. Great news for humans. Two caveats: SemiAnalysis built the test with outside partner companies and grades its own rigor, and safety filters blocked some models from cyber and biology tasks, which pulled their scores down. If you're still sorting out which AI names to try, our guide to every AI model worth knowing compares more than 30 of them and points out the free ones.
JOIN THE COMMUNITY
Learn AI With Us, Free
Most people try AI once, get a so-so answer, and quit. Our free 14-day challenge gets you past that point, one short lesson a day.
Week one, AI starts working for you. It drafts a real email for you, flags which messages in your inbox need a reply, and turns messy notes into next steps. You also learn to catch its mistakes. By day 7 you have a weekly AI routine you'll actually keep.
Week two, you build something of your own. You find a skill people will pay for and turn it into an offer, with a one-page pitch, a short video and your first client emails, all drafted with AI and ready to test. Day 14 hands you a 30-day plan.
10 to 15 minutes a day. No tech background needed. There's even a cinematography course coming out soon.
QUICK TAKES
The story: OpenAI emailed desktop app users that the Chat and Work tabs are becoming one, so you can go from an idea to having ChatGPT build a website or a slide deck in the same place. ChatGPT and Codex will also start sharing one usage allowance, with a test group getting it first.
Your takeaway: One tab is simpler. OpenAI admits some people may hit limits and wait for resets more often, so heavy Codex users should keep an eye on their allowance.
The story: Tutor Intelligence raised a $75 million Series B led by Andreessen Horowitz, bringing its total to $117 million. Remote human operators guide its warehouse robots, Cassie and Sonny, during real shifts, and every finished task becomes training data.
Your takeaway: It's today's lead from the other direction: machines get better by watching people do the work. Tutor says its robots now run at nearly four times as many sites as a year ago. For more on how robots and drones learn, see our robotics and drones hub.
The story: Anthropic is extending its text watermark, a hidden pattern that detection tools can spot, to older Claude models worldwide with no opt-out, The Independent reports. xAI says its Grok models will add one by default on November 7, and both companies point to EU AI Act rules.
Your takeaway: You can't check the mark yourself, since only approved groups like regulators and researchers get detection access. Anthropic also says short texts may not carry enough of the pattern to be caught, so think of it as a trace that sometimes fails.
TOOLS ON OUR RADAR
📋 Plane Free and Open Source: Track projects and tasks for up to 12 people free, a stand-in for Jira or Asana.
📄 PDFgear Free: Edit, convert and chat with PDFs with no sign-up, an Acrobat stand-in for checking the contracts AI still fumbles.
💽 WizTree Free: Shows in seconds which files are filling your Windows or Mac drive, free for personal use.
✍️ DocuSeal Free and Open Source: Send 10 e-signature requests a month free, a DocuSign alternative for leases and forms.
🎙️ Riverside Freemium: Record podcast interviews in your browser; the free plan caps video at 720p and adds a watermark.
🧠 Flowith Freemium: An AI canvas for multi-step research, with 300 one-time free credits, then $19.90 a month.
🗣️ BetterDictation Paid: Type by talking in any app on Apple Silicon Macs, offline, for a one-time $39.
TRENDING
Musk Tightened Grok Bot Limits Over "Unsustainably High" Costs. Some subscribers watched their weekly allowance jump from about half used to maxed out over Halloween weekend, and Musk confirmed on X it wasn't a glitch. He hasn't said what the new limits are.
Squarespace Is Cutting About 300 Jobs. That's 19% of its staff. The company calls it a restructuring, and it comes as AI tools that build a website from a single prompt crowd into its business.
A Startup Says Its AI Can Predict How You'll Trade. Israel's 42AI came out of stealth with $62.5 million, mostly from its founders' own investment group, and works with 10 brokerages. The accuracy figures are its own, and a co-founder called misuse "a valid concern."
Glasses That Turn Silent Mouthing Into Text. Subvocal's glasses send radio waves through your face, and AI decodes the words you mouth without a sound. The first 500 preorders ship in early 2027, and Hacker News commenters asked who wants to mouth at their laptop in public.
Hume's New AI Listens for Frustration in Your Voice. Its Audio Decision API answers questions about a recording, like whether a caller sounds annoyed. In Hume's own test it matched a human panel's judgment 70.7% of the time, versus 56.5% for a single person.
TRY THIS PROMPT (copy and paste into Claude, Grok, ChatGPT, or Gemini)
🧭 Hand It Off? Sort any task into what AI can do and what still needs you
Build a single-file HTML app called Hand It Off? in vanilla HTML, CSS and JavaScript. No API key needed.
I describe a task I want AI to do, like reading my lease to find early move-out fees, and pick how bad a wrong answer would be: low, medium or high.
The app calls the Anthropic Messages API to split the task into 3 to 6 steps and sort each one: AI can draft it, AI helps but I verify specific things, or keep it human. Higher stakes mean more caution. It must never invent statistics, prices or laws, and for legal, medical or money questions it names the kind of professional to ask instead of giving advice.
Show a three-color bar for the split, one card per step with a one-line reason and a checklist of what to verify, and a short watch-for list. Style: near-black background, teal for AI, orange for checks, pink for human, Space Grotesk and Newsreader fonts.What this does: Describe a task you'd like AI to take over and say how bad a wrong answer would be. The app breaks it into steps and tags each one: AI can draft it, AI helps but you check it, or keep it with a person, plus a short list of exactly what to verify. On a day when the best AI finished 12% of real expert tasks, it's a quick way to see where your own eyes still matter.
15,000 Books. Learn From Them in Minutes.
From Atomic Habits to Never Split the Difference, Shortform gives you the key ideas and how to use them in minutes, not months.
WHERE WE STAND(based on today’s news)
✅ AI Can Now: Work for hours on its own inside a 2,000-file deal room, real lab data or a chip design task.
❌ Still Can't: Finish most of that work. The best model completed about 12% of GDPx tasks, and nearly every model scored zero on pass-or-fail finance and legal tasks.
✅ AI Can Now: Hide a pattern in the text it writes so approved groups can tell it came from Claude, and soon Grok.
❌ Still Can't: Catch every case. Anthropic says short texts may not hold enough of the pattern, and critics say rewording with another AI can wash it out.
FROM THE WEB
RECOMMENDED LISTENING/READING/WATCHING
🎬 Humans Need Not Apply by CGP Grey (2014) - Video
This 15-minute video argued back in 2014 that bots would come for white-collar jobs too, lawyers and doctors included, and it's been watched more than 18 million times. Watch it next to today's lead. Twelve years later the best AI finishes about 1 in 8 real expert tasks, so the prediction still stands, on a slower clock than the video suggested.
How was today's edition? Tap one below, then tell me what you'd change on the next screen. I read every answer.
How was today's edition?
Thank you for reading. We’re all beginners in something. With that in mind, your questions and feedback are always welcome and I read every single email!
-James
By the way, this is the link if you liked the content and want to share with a friend.
Some * designated product links may be affiliate or referral links. As an Amazon Associate, I earn from qualifying purchases. This helps support the newsletter at no extra cost to you and Amazon makes a tiny hair less.
