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!
A video-call AI just fooled about half the people who talked to it in a one-minute test. The company that built it says it isn't safe to release yet, and I think that's the most useful part of the story. I'll cover what was tested, what it means for the next "urgent" call you get, and one habit that protects you. As always, every story here is human picked and edited.
The AI Quick Score (free)product for websites will be coming out in the next couple days. It's due for an update. This will be version two that gets released, and will take into account the many updates and changes that have taken place with how search engines and AI chats currently operate.
THE FRONT PAGE
An AI Fooled 48% of People on a Live Video Call, and It Isn't Out Yet
TLDR: Tavus says its new video AI, Griffin, made 26 of 54 people in a blind test believe they were on a live video call with a real person.
The Story:
Tavus, a San Francisco startup, calls Griffin a "human interaction model." Instead of chaining speech recognition, a chatbot and a talking-head generator together, it sees, listens and responds in one model. It watches your face and gestures, can nod along or interrupt, and starts answering in 0.43 seconds, according to the company.
In the test, 54 people recruited through an outside platform each had a one-minute video call they thought was with a human. Afterward, 48% said they'd talked to a real person, and they were about 79% sure of it. Tavus's previous system fooled 1 of 41 people.
Its Significance:
A 48% score is close to a coin flip, so people in the test couldn't reliably tell the difference. Crypto Briefing points out the limits: 54 people is a small sample, a minute is a short call, and Tavus ran the test itself. Tavus is keeping Griffin to a small group of testers and says it needs more safety work first, because realism that feels natural can also fool people. Set up a family code word now, and treat any video call asking for money or passwords as unconfirmed until you call back on a number you already have. And when it comes to your company, consider upgrading any protocol that deals in sensitive information or finances being handled over video calls.
QUICK TAKES
The story: Apple says some apps use macOS's Full Disk Access setting in ways that put people at risk, and the danger grows as AI agents get more capable. In macOS 27, granting it will take what Apple calls "very explicit user action."
Your takeaway: Full Disk Access lets an app read your files, mail, messages and browsing history. Open System Settings, then Privacy & Security, then Full Disk Access, and switch off anything you don't recognize.
The story: Meta released free, open-source firmware and a Linux toolkit so hobbyists can connect Muse to a Raspberry Pi, an ESP32 board or their own device with a screen, buttons or sensors. It's also giving 5,000 free Home Link plugs, which let Muse talk to your TV and speakers, to Muse subscribers.
Your takeaway: Meta's starter ideas include an e-ink display and a TV stick, so a weekend tinkerer can build a Muse screen for the kitchen. The free Home Links are first come, first served. If you haven't tried it yet, their freemium level has a very high usage limit. You can get a billion(yes with a b) additional Muse credits with my invite code for as long as it lasts, OK2I3S My first impression is very good, and a comparison video will be coming out on YouTube soon.
The story: Suno Speech, now in open beta on the web and in Suno's apps, turns text into spoken audio with original background music in one track. You pick the voice and the musical style.
Your takeaway: Suno's own examples include bedtime stories over soft piano and hype speeches over stadium drums. It's a quick way to make a podcast intro or a birthday message, though Suno warns accents can drift during the beta.
TOOLS ON OUR RADAR
🔐 Ente Auth Free and Open Source: Two-step login codes with encrypted backups, a free swap for Authy and Google Authenticator.
⏱️ ActivityWatch Free and Open Source: Shows which apps and sites eat your day, keeping the data on your computer instead of RescueTime's cloud.
🚧 LeechBlock NG Free and Open Source: A browser add-on that blocks distracting sites on a schedule you set, a free Freedom alternative.
✅ Tasks.org Free and Open Source: An ad-free Android to-do app with subtasks, tags and Google Tasks sync, a free Todoist swap.
📬 Thunderbird Free and Open Source: Mozilla's email app puts all your inboxes, calendar and contacts in one ad-free place, like Outlook.
🗣️ TalkPal* Freemium: Practice a new language out loud with an AI tutor, free for 10 minutes a day.
🎞️ CapCut* Freemium: A free video editor with auto captions and templates for Reels and TikToks; Pro adds more AI tools.
TRENDING
Nvidia's Cheaper Desktop AI Computer Costs $4,999. The 64GB DGX Spark arrives October 23 from six makers including Dell, HP and ASUS, and runs models up to 100 billion parameters. Rising memory prices pushed the 128GB version to $6,950.
Anthropic Puts $100 Million Into Training 10,000 AI Engineers. Claude Frontier Academy copies medical residencies: in-person training, then 12 weeks leading a real Claude project at work. It's nomination-only through employers such as Deloitte and Morgan Stanley, with first credentials due early 2027.
Napster's Sean Parker Is Turning Stability AI Into a Music Company. The image-generator maker took $76 million from Sony, Warner and Universal in August, plus licenses to train on their catalogs. Parker says a coming feature will let you hum a melody to steer a song.
OpenAI's Head of Safety Transparency Reportedly Quit. David Robinson oversaw system cards, the documents that explain a model's risks. It follows the firing of three safety researchers accused of sharing data with an outside group; OpenAI hasn't commented.
Chinese Hackers Posed as an Anthropic Employee to Phish AI Policy Experts. Proofpoint says the group, TA419, also impersonated a former White House official, invited targets onto fake committees, then sent them to a fake OneDrive login that grabbed passwords and one-time codes.
TRY THIS PROMPT (copy and paste into Claude, Grok, ChatGPT, or Gemini)
🛑 Pause Before You Pay Describe an urgent call asking for money or a code, and get a calm plan to check if it's real.
Build a single-file HTML app called Pause Before You Pay in vanilla HTML, CSS and JavaScript. No API key needed.
I pick who the caller said they were, how they reached me (video call, phone, text or email), and tick what they asked for: money, gift cards, crypto, a login code, a password, installing an app, keeping it secret, or acting right now. I can also describe what happened.
The app scores common warning signs itself, then calls the Anthropic Messages API to write a calm plan for checking the request through a number or app I already trust. It must never claim a call was or wasn't a deepfake or a scam, and never invent facts.
Show a warning-sign meter with the reasons, numbered steps, questions only the real person could answer, and a short list of what not to do. Style: near-black background, coral and mint accents, Space Grotesk and Newsreader fonts.What this does: Pick who the caller claimed to be, how they reached you and what they wanted. The app flags the warning signs, then gives you step-by-step ways to confirm it's them, plus questions only the real person could answer. It won't pretend it can spot a deepfake; it helps you slow down and check.
What’s still sitting on your to-do list?
You don’t need an AI team to build an AI agent. With Skydive, the person who knows the work can create the agent to do it.
Start an agent off with a role, teach them how your team operates, connect your tools, and put them to work.
WHERE WE STAND(based on today’s news)
✅ AI Can Now: Hold a live video call that about half of people in a one-minute test mistook for a real person.
❌ Still Can't: Show it holds up longer. Tavus tested 54 people for one minute each, with no outside review.
✅ AI Can Now: Read a script aloud and compose matching background music in one pass.
❌ Still Can't: Keep a voice steady. Suno says accents can drift mid-track during its beta.
FROM THE WEB
RECOMMENDED LISTENING/READING/WATCHING
🎬 Ex Machina (2014) - Movie
Alex Garland's thriller sends a young programmer to a remote house to test whether a humanoid AI named Ava can pass as conscious. It's a Turing test played face to face, the same question Tavus asked this week, with far higher stakes and no one-minute limit. Quiet, tense, and always asking who is testing whom. It won the Oscar for visual effects.
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.
