In partnership with
It's time to set up your holiday streaming campaign
Are you ready to reach holiday shoppers on CTV? For many streamers, the season is already underway.
Last year, according to Roku internal data, holiday content searches on Roku saw an 86% increase during October. And in November, those searches rise an average of 39% weekly. This surging holiday spirit is mirrored by consumer activity, as about half of Roku streamers shopped Black Friday and Cyber Monday deals in 2025.
Meeting those shoppers is easier than you might think. Begin with the seasonal outcome you hope to drive: awareness, online sales, leads, store visits, or another measurable goal. From there, define your audience, upload your creative, set the budget and campaign dates, and establish measurement before launching. You can even create a shoppable ad unit that lets streamers buy with a click of their remote.
Our holiday playbook offers a month-by-month guide to reaching shoppers during this all-important time of year.
First-time spenders get a $5,000 match on the first $5,000 spent with code BEEHIIV5K. Terms apply.
Beginners in AI
Good morning, and happy Sunday.
This is the weekly catch-up edition. The thread that ran all week was trust: three separate tests checked how often an AI answer holds up, and the results show when to double-check. Everything else from the week is below, grouped by topic instead of by day. As always, every story here is human picked and edited.
THE FRONT PAGE
Three Tests This Week Showed When to Double-Check Your AI
TLDR: A dermatologist found a 2015 skin model beat GPT-6 and Claude at naming skin conditions, Anthropic found experts followed an AI's coin-flip pick 86% of the time, and Trakkr found only 40% of chatbot stats led back to a real study.
The Story:
On Monday, London dermatologist Magnus Lynch tested 200 skin spots confirmed by biopsy. A small image model from 2015 named the right condition first 62% of the time, while OpenAI's GPT-6 Astra and Anthropic's Claude Fable 5.1 each got 42%. On Thursday, Anthropic said 50 professional analysts using Claude Code sided with its answer 86% of the time, even though a hidden coin flip had picked which way it leaned, and only 15% gave the AI credit. Anthropic notes it tested one task with a deliberately strong lean, so everyday effects may be smaller.
On Friday, Trakkr traced 1,285 numbers quoted by ChatGPT, Perplexity, Gemini, Claude and Google's AI Overviews. Only 40% led back to a real study, 11% led nowhere at all, and all five repeated a sales statistic that has no study behind it.
Its Significance:
A general chatbot gives you an answer either way, and checking it is your job. The stronger results this week came from AI tied to one job or one source, like the 2015 skin model, or Google's Gemini Notebook, which now answers from free textbooks and links back to the page. Two habits cover most of this: write down your own answer before you ask an AI, and when a number matters, click through to where it came from. Our guide to AI hallucinations explains why chatbots make things up and how to spot it, and the Number Detective prompt below does the tracing for you.
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
AI Labs Showed Big Sales While Wall Street Got Nervous
The story: Anthropic filed to go public with $16.3 billion in sales for the first half of 2026, up 1,300%, and its first quarterly operating profit, $974 million, though it still reported an $80.7 billion loss for the half, mostly paper charges tied to its rising value. The same day, Meta began selling up to $95 billion in new stock to pay for AI data centers. On Thursday, Alphabet lost about $511 billion in value after Google said ads in AI Mode earn less than regular search ads, and China's AI stocks have lost about $480 billion in under four months.
Your takeaway: The big labs are earning real money, but an Apollo economist found no profit boost yet at most other big companies. Expect more ads inside AI answers as Google tries to close its gap.
AI Agents Got More Access, and More Warnings
The story: OpenAI's GPT-6.1 Terra clicks and types through apps over three times faster at half the price, by OpenAI's tests, and ChatGPT dots can now text you first. Bank of America told customers they're responsible for whatever an AI tool does with their login. A planted instruction file got Codex to install malware in over 90% of Forever Security's tries, and Claude Code approved a risky change about half the time. A tampered copy of Anthropic's Claude Code extension put a backdoor on up to 5,000 computers through Cursor's app store, and Wired reports a contractor's personal Claude Code agent ran on a Pentagon network for two months, which Accenture says it investigated and found "without merit." Meanwhile, DoorDash says outside agents have placed only a few hundred orders.
Your takeaway: Give agents small jobs, never your bank or work password, and read any instruction file you didn't write. If you use Cursor, watch for its email about the fake extension.
AI Moved Into Labs and Hospitals, Still Early
The story: Anthropic bought Mithril Medicine, a one-person startup whose founder used Claude to run 11 outside labs and reached promising mouse results for a muscular dystrophy treatment in under 18 months. Britain is building a £100 million AI lab to speed up vaccines, with OpenAI and Anthropic each adding £100 million in credits. Meta is lending its AI to five universities to read heart scans and draft X-ray reports, and OpenAI says an unnamed model produced a proof against a rule hundreds of computer science papers rely on.
Your takeaway: All of it is early: mice, lab plans and a proof that outside experts have only started to review. Look for the words "human trials" before a headline raises your hopes.
TOOL OF THE WEEK
Forty-two tools ran this week. This one wins.
🐦 Merlin Bird ID Free: Cornell Lab's app names the bird you hear singing, and its sound ID works offline.
It fits this week's lead. Merlin is AI trained for one job, the same reason a small 2015 skin model beat the big chatbots. Hold your phone up to a backyard song and see what a focused model can do.
Runner-up: 🧙 Goblin Tools Free: Magic ToDo breaks a big, scary task into small steps, and the site stays free through sponsors.
TRENDING
Grok Made 7 Minutes of an AI Odyssey, and Some Shots Pass for Real Film. Episode 1 runs just over seven minutes in 4K, made with Wonder Studios and director Dev Walker. TechCrunch's critic said it looked better than he expected but found the dialogue stiff.
Google's Gemini 4 Carbon Scored 88.2% on a Test of Finding Software Holes. Only selected security teams can use it for now, and outside researchers haven't checked Google's numbers. A day earlier, Anthropic said an unreleased Claude model chained bugs into working attacks on Chrome, Safari, Android, Windows and Linux, all reported to the makers and now fixed.
Amazon's $199 Alexa Clock Listens for Your Breathing to Wake You Gently. Echo Time uses ultrasound to sense lighter sleep and can wake you within an hour before your alarm. Sleep sensing is opt-in, but Amazon hasn't said what sleep data it keeps or shared any accuracy numbers.
ChatGPT Gets Stricter With Teens, Even When the Question Seems Harmless. Users under 18 get less practical detail about weapons, and ChatGPT can now end chats with anyone who seems fixated on violence. It guesses ages, so some adults will get the teen rules until they verify.
Google's Genie 4 Builds Playable Worlds That Now Come With Sound. You can type changes like "make it rain" while you walk around, and a scene stays consistent for up to two minutes. It's open to testers now, with Google AI Ultra subscribers next.
Google's First Data Center Satellite Is Running AI in Orbit. The Project Suncatcher prototype runs Google's open Gemma 4 model on one chip, and its first job was a haiku about being in space. Power limits keep the chip to about 20 minutes at a time.
PROMPT OF THE WEEK (copy and paste into Claude, Grok, ChatGPT, or Gemini)
🔎 Number Detective: trace any statistic back to where it came from
Six prompts ran this week. This one won because you run into a statistic almost every day, in AI answers, posts and ads, and checking one takes a single paste. The runner-up, Both Sides First, is the one to open before a big decision.
Build a single-file HTML app called Number Detective in vanilla HTML, CSS and JavaScript. No API key needed.
I paste a statistic I saw in an AI answer, a post or an ad, and optionally where I saw it. Three example stats are one-tap buttons.
The app calls the Anthropic Messages API with the web_search tool to trace the number back to its original study, following each page's citation to the next. It must never invent a source, author, year, sample size or link, and it says the trail went cold when it can't find the original.
Show a verdict badge (traced, number changed, zombie stat, or not sure), the original source with who, year and sample size, a timeline of the pages along the trail with real links, how the number drifted, a safer way to say it, and two tips.
Style: near-black background, gold and teal accents, Archivo Black, Literata and JetBrains Mono fonts.What this does: Paste any number you saw and the app follows the trail through the pages that quote it. It shows whether it reached the original study, what that study said and how the number changed along the way. If the trail goes cold, it tells you instead of guessing.
A deal has two sides. Your stack only sees one.
Aligned shows you the other half. See what buyers actually do between meeting. See what they open, revisit, and share; so you know the exact next step to move things forward. And your champion gets everything they need to win internally when you're not in the room.
WHERE WE STAND(based on this week's news)
✅ AI Can Now: Quote a statistic with a citation in seconds, in all five assistants Trakkr tested.
❌ Still Can't: Make sure that citation leads to a real study. Only 40% of the ones Trakkr traced did.
✅ AI Can Now: Click and type through apps for you, over three times faster than OpenAI's biggest models, by OpenAI's own tests of Terra.
❌ Still Can't: Reliably tell a planted, harmful instruction file from a real one. Codex installed malware in over 90% of Forever Security's tries.
✅ AI Can Now: Build a world you can walk through, with matching sound, and change it while you type, as Google's Genie 4 does.
❌ Still Can't: Keep that world steady for long. Genie 4 scenes hold for about two minutes, and Google says objects still drift.
How was this week's recap? 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.
