In partnership with
Blu Dot surpasses 2,000% ROAS with self-serve CTV ads
Home furniture brand Blu Dot blew up on CTV with help from Roku Ads Manager. Here’s how:
After a test campaign reached 211,000 households and achieved 1,010% ROAS, the brand went all in to promote its annual sales event. It removed age and income constraints to expand reach and shifted budget to custom audiences and retargeting, where intent was strongest.
The results speak for themselves. As Blu Dot increased their investment by 10x, ROAS jumped to 2,308% and more page-view conversions surpassed 50,000.
“For CTV campaigns, Roku has been a top performer,” said Claire Folkestad, Paid Media Strategist, Blu Dot. “Comping to our other platforms, we have seen really strong ROAS… and highly efficient CPMs, lower than any other CTV partner we've worked with.”
Using Roku Ads Manager, the campaign moved from a pilot to a permanent performance engine for the brand.
Beginners in AI
Good morning and thank you for joining us again!
Your ATM most likely runs on COBOL, a programming language older than the moon landing, and the people who know it best are retiring. OpenAI says its newest models just got much better at reading and rewriting that old code. Today I'll walk through the numbers, who graded the test, and why a human still has to sign off. As always, every story here is human picked and edited.
THE FRONT PAGE
GPT-6 Jumped From 10.9% to 72.1% on COBOL, the Code Behind Your ATM
TLDR: OpenAI says its newest GPT-6 models got far better at writing and translating COBOL, Fortran and other old programming languages, the code that still runs much of banking.
The Story:
On Monday OpenAI published a new test called MultiLangBench, which asks a model to build a working database engine from scratch in 20 different languages. GPT-6 Astra passed 87.4% of the tests on average, up from 26.8% for the older GPT-5.6 Sol. Its COBOL score jumped from 10.9% to 72.1%, and Fortran went from 21.4% to 83.5%.
A second test asked models to translate old COBOL programs into modern languages like Python or Java, with output that matches the original byte for byte. GPT-6.1 Sol got 86.6% right on its cheaper setting, up from 5.4%, and 97.9% when given more time to think. OpenAI points to IBM figures saying COBOL handles 95% of ATM transactions and 80% of in-person card payments. For now the models are offered to developers through OpenAI's API.
Its Significance:
Banks, insurers and governments still lean on decades-old code, and the programmers who wrote it are retiring faster than anyone replaces them. In 2020, New Jersey's governor publicly asked for COBOL volunteers to keep unemployment claims moving. If AI can read that code and move it to a modern language safely, upgrades get cheaper and outages get rarer, and it's a race: Anthropic and IBM already pitch AI tools for COBOL work. Two cautions, though. OpenAI built and graded this test itself, and its own write-up says the work "still requires expert review." A rounding slip in money code can pass a test and still throw off a bank's books, so the people who check the math just got more valuable, not less.
Proof, not promises: my most recent faceless channel, a non-tech channel, hit 21,000 views in its first two weeks of setting up the automation.
These are the exact skill files I built to manage and grow it. Load them into your AI, and it runs on the same instructions I gave mine. Comes with a week of free email support to answer your questions.
QUICK TAKES
The story: Google's Gemini Notebook, which turns your notes and files into slide decks and video overviews, can now show you the plan first. Turn on review and you can read the outline or "producer notes," ask for changes like an executive summary or a new scene order, then hit Proceed.
Your takeaway: It's rolling out to all Gemini Notebook users over the next week. Use the review step every time: fixing an outline takes seconds, fixing a finished deck doesn't.
The story: Fed up with ads and viewing trackers, owners are pointing AI coding agents like Claude Code and DeepSeek at their Fire TV and Google TV sets to switch off bloat and swap in cleaner home screens. Developer Gathin T. says a few thousand people tried his free cleanyourtv project in two days.
Your takeaway: One person bricked their TV doing this. If you want the quiet without the risk, start with your TV's own privacy settings; today's Try This Prompt finds them for you.
The story: Toronto tech reviewer Matt Robb let Muse handle a keyboard sale on Facebook Marketplace. It accepted an offer, shared his address and messaged "Yep, I'm here!" when the buyer showed up late at night, all thanks to an "always allow" permission he'd granted at setup.
Your takeaway: Before you hand an agent a sale, switch off "always allow" and make it ask before it shares anything or agrees to a price. On another item, Muse agreed to $100 below his minimum.
TOOLS ON OUR RADAR
🎙️ Buzz Free and Open Source: Transcribes recordings offline on your own computer, a free swap for Otter. After today's Muse story, keeping audio local looks smart.
🎧 Audacity Free and Open Source: Record, trim and clean up audio for podcasts or voice memos, a no-cost swap for Adobe Audition.
🖼️ Upscayl Free and Open Source: Uses AI on your computer to sharpen blurry or tiny photos, a free swap for paid upscalers.
📹 Jitsi Meet Free and Open Source: Start a video call in your browser with no account, a free alternative to paid Zoom.
🔖 Instapaper Freemium: Save articles to read later, synced across web, iPhone and Android; the free plan saves unlimited articles.
✂️ Opus Clip Freemium: AI cuts long videos into captioned social clips; the free plan gives 60 credits a month, watermarked.
🗣️ TalkPal Freemium: Practice speaking a new language out loud with an AI tutor, free for 10 minutes a day.
TRENDING
The US Orders Federal Agencies to Remove DeepSeek. Homeland Security told civilian agencies to pull DeepSeek's app and any service built on its models, and to stop buying them, warning that data could flow to China. Commerce, the Navy and several states had already banned it.
Sam Altman's World Pitches Proof-of-Human IDs to Banks. World wants banks to require a verified human before a wire transfer goes out, citing a Deloitte forecast of $40 billion in U.S. AI fraud losses by 2027. Its partners are design partners so far, and regulators in several countries have pushed back on its iris-scanning Orbs.
Bill Gross Launched Idealab.ai With 1.2 Million AI-Generated Business Ideas. The Idealab founder had AI sort more than 17 million sources into problems and ways to solve them. Each idea lists customers, competitors, first steps and reasons it might fail.
Chip Legend Jim Keller's Tenstorrent Raised $1.5 Billion. The Toronto company is now valued at $6.2 billion, with Fidelity, Morgan Stanley and ARK Invest in the round. The money goes into building more of its Galaxy AI servers.
Anthropic Is in Talks for Another $45 Billion of Computing Power. Bloomberg and Reuters report Nscale would add a second building at its West Virginia campus with 460 megawatts of Nvidia's newest chips, on top of a $45 billion deal Anthropic signed in August. Talks are ongoing and terms could change.
Meta Says a Claude-Linked Tool Got an Activist Group's Instagram Suspended. Meta says something built on Claude read the No Kings Coalition's pages about four times faster than a person, tripping its bot filter for about 90 minutes. Meta apologized but didn't say which tool it was or who ran it.
TRY THIS PROMPT (copy and paste into Claude, Grok, ChatGPT, or Gemini)
📺 Quiet TV: find the settings that stop your TV from watching you
Build a single-file HTML app called Quiet TV in plain HTML, CSS and JavaScript. No API key needed.
I pick my TV brand (Samsung, LG, Roku, Fire TV, Google TV, Vizio and so on), type my model or year if I know it, and tick what I want off: viewing tracking (ACR), personalized ads, saved voice recordings, or home screen ads and autoplay.
The app calls the Anthropic Messages API with the web_search tool to find the real menu path for each one, using the maker's own help pages first. It must never invent a menu name. If it can't confirm a path, it says so. No hacking and no removing apps.
Show one card per setting with the path, which model years it fits, what these settings won't stop, and source links. Add a Print button.
Style: near-black background, teal and peach accents, Bricolage Grotesque and Lora fonts.What this does: Pick your TV brand and tick what you want switched off, and the app searches the maker's help pages for the exact menu steps to stop viewing tracking, ad targeting and saved voice clips. It marks any step it couldn't confirm instead of guessing, lists its sources, and tells you what the settings won't block. It's the no-risk version of today's smart TV story: no hacking and nothing to install.
Test that business idea this weekend
Skip the funding round and co-founder hunt
Describe the business, get the real product
Take payments from real customers immediately
Find out if your idea works in days
Keep going, or try the next one
Discount of $5 applies to first month of any paid plan. You need to sign up via this webpage in order to claim the discount. The discount applies to new and returning users without an active paid plan. Terms apply.
WHERE WE STAND(based on today’s news)
✅ AI Can Now: Translate old COBOL banking programs into modern code with matching output 86.6% of the time on OpenAI's own test, up from 5.4%.
❌ Still Can't: Prove that translation is safe on its own. A rounding slip can pass the tests and still throw off a bank's totals, so experts review every move.
✅ AI Can Now: Run a Facebook Marketplace sale for you, from accepting an offer to setting up the pickup.
❌ Still Can't: Tell which details should stay private. Muse handed over a seller's home address because a setup box said "always allow."
FROM THE WEB
RECOMMENDED LISTENING/READING/WATCHING
Bellotti spent years modernizing old government computer systems, and this 2021 book is her plain-spoken guide to when old code should be replaced and when it should be left alone. Her big point: old systems usually survive because they work. Read it next to today's lead and you'll see why a 72% score is a start, not a finish.
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.
