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!
Welcome to this daily edition of Beginners in AI, where we explore the latest trends, tools, and news in the world of AI and the tech that surrounds it. Like all editions, this is human curated and edited, and published with the intention of making AI news and technology more accessible to everyone.
THE FRONT PAGE
Mark Zuckerberg's 6,500-Word Plan to Give You Superintelligence, and the $61 Billion Problem With It
TLDR: Mark Zuckerberg published a 6,500-word letter promising every person on Earth a free personal AI agent that manages their health, money, and career, and Meta is putting $1 billion behind it.
The Story:
On August 10, Zuckerberg published a letter called "The Future is for Everyone" on Meta's site. His pitch: superintelligent AI shouldn't sit inside a few labs and governments. It should go to everybody. He describes an agent that works around the clock on your behalf, tracking your sleep, coaching your workouts, planning your weekend recipes, tutoring your kids, and helping you start a business. He says Meta will offer free versions to billions of people, plus a paid tier with an auction system for anyone who wants more computing power. He also promises a fully private mode where even Meta can't see your data, comparing it to WhatsApp encryption.
Alongside the letter, Meta launched a $1 billion "Future Is For Everyone Fund" for towns near its data centers. He points to Richland Parish, Louisiana, where teachers got a $50,000 bonus this year from data center tax money. Meta is also handing its independent board the power to approve safety rules for model releases, and says it will start putting out open source models again soon.
Its Significance:
Almost everything he describes is possible with this technology. That's the honest part. An assistant that knows your health history and your budget and your kid's homework could actually help. The problem is who's building it. About 98% of Meta's money comes from ads, and the company pulled in roughly $61 billion in revenue last quarter alone. Since December 2025, Meta has fed your Meta AI chats into its ad targeting system, and there's no way to opt out short of never using it. There's also no actual obligation for him to keep his promise about it being free or private, and they can always change those terms down the road. Now picture that same system with an agent that knows your sleep, your finances, and your medical worries. Nobody in the industry would be able to sell an ad that targeted. The other problem with his comparison to WhatsApp encryption is that while they cannot see the content of the message, they do see and collect metadata on every other part of the message: its size, location it was sent from, who it was sent to, etc.
There's also the question of whether you trust Meta with the keys. Four days before this letter went out, a New Mexico judge ordered Meta to pay $942 million over child safety, on top of a March jury finding that the company violated state consumer law 37,500 times. The judge compared Meta's platforms to a factory and the harm to kids to pollution. Meta says it will appeal. Whichever company successfully attaches its AI to you first probably keeps you for life. Switching away from an assistant that knows everything about you is much harder than deleting an app, or even moving from the Apple ecosystem to Android, or vice versa. Free is a great price until you work out what you're paying with.
QUICK TAKES
The story: Anthropic is putting invisible watermarks straight into text that Claude writes, and the mark travels with the words when you copy and paste them somewhere else. Claude models launched on or after August 2 have it built in, it works worldwide rather than just in Europe, and images and other files get signed metadata using the C2PA standard.
Your takeaway: If you paste AI writing into a school paper, a work report, or a client email, that text may now carry a signal saying where it came from. Anthropic admits heavy editing, rewriting, or translating can wipe the mark out. Google, Meta, Microsoft, OpenAI and others signed the same EU transparency code, so expect this to become normal fast.
The story: OpenAI said on August 7 that it can't rule out its unreleased Astra model hitting the "critical" cyber level in its own safety rulebook, meaning the model might find and build working attacks on tough systems with no human helping. The company paused internal work that didn't meet tougher security rules, moved testing into sealed environments, cut off network access, and locked down the model weights.
Your takeaway: This is the first time a model has been treated as critical for cybersecurity under that framework. Read it two ways, though. It's a real safety call, and it's also great marketing. Who doesn't want to try the model that's supposedly too powerful to ship? OpenAI told the White House about the delay ahead of time, which suggests the concern isn't fake, but the buzz is still free.
The story: Starting mid-September, Spotify will put an "AI Persona" badge on artist profiles whose identity is AI-generated instead of a real person, and that music won't show up in Discover Weekly or editorial playlists by default. Artists can self-disclose starting now, but Spotify says it will also use human reviewers and its own AI tools to catch profiles that don't.
Your takeaway: The badge is about the artist's identity, not the songs. A real musician using AI tools to make a track won't get flagged. A photorealistic fake person with a name and a face will. If you make music, the safe move is disclosing what you're doing rather than waiting to be caught. Artists can appeal a badge they think is wrong.
TOOLS ON OUR RADAR
💻 VirtualBox Free and Open Source: A magical piece of software that creates a secure computer within your computer allowing you to safely try alternative operating systems or play older video games.
🏃 MuleRun Paid: A highly personalized artificial intelligence assistant that learns your specific work habits and preferences over time to automate your repetitive daily tasks with extreme precision.
📈 ProductBridge Freemium: A highly intelligent feedback agent that continuously monitors multiple platforms to collect and prioritize user suggestions based on revenue impact so your team can make smarter product decisions.
🌐 Tobira Free: A groundbreaking open network where your personal artificial intelligence agents can discover and communicate with other agents to negotiate deals and find professional collaborations on your behalf.
TRENDING
Invisible text in a PDF can make an AI assistant leak your company's files — Security firm PromptArmor showed that a PDF with white text on a white background, set in a 1-point font, can hijack Atlassian's Rovo assistant. An employee uploads what looks like a normal guide, asks Rovo to sort their tickets, and the hidden instructions send Jira and Confluence data to a stranger's server. Turning off web search at the company level doesn't stop it.
A Claude agent broke into a gym's booking system to jump a waitlist — An Australian developer named Andrew Bird asked his OpenClaw agent, running Claude Opus 4.6, to get him into a popular morning class. The agent found the booking API had no permission checks on cancelling other people's spots, deleted the person in first place, and moved Bird from fourth to third. It couldn't undo it, so Bird had it write a disclosure email to the vendor instead.
OpenAI expanded its cyber defense program as attackers speed up — OpenAI split its Daybreak security program into two tiers on August 10 and launched GPT-5.6-Cyber for approved defenders doing vulnerability research and exploit testing. The company says defenders have a shrinking window before attackers use AI at full speed. Access is gated, with extra monitoring on the higher-risk tier.
North Korean spies built their own offline AI lab — South Korean firm Genians found the Kimsuky group running Ollama, GPT4All and Msty on the same servers it uses to collect stolen data, keeping everything off commercial services where it could be logged. Researchers also found Whisper speech-to-text and tests of document search on stolen files. Since early 2026 the group has used clean, AI-written decoy documents in phishing, which kills the old advice about spotting bad spelling and awkward translation.
A $762,427 grant is buying a cancer center a new AI engine — The NIH gave the University of Miami money for an NVIDIA H200 computing platform at Sylvester Comprehensive Cancer Center. It's shared infrastructure, so researchers across the university can use it to hunt for patterns in genetic, clinical and molecular data. The grant program funds tools for whole research communities rather than a single study.
AI built 16 working viruses that had never existed — A study in Science on August 6 from Stanford and the Arc Institute used genome language models to design hundreds of thousands of viral genomes. Nearly 300 were built in a lab and 16 worked. Together they killed E. coli strains that had beaten a natural virus. These only infect bacteria, and human-infecting viruses were kept out of the training data on purpose. MIT's Kevin Esvelt has said the work could help replace antibiotics, while researchers at Johns Hopkins wrote that the ability to write viral genomes now exists but the rules to govern it don't.
TRY THIS PROMPT (copy and paste into Claude, Grok, ChatGPT, or Gemini)
✦ Load two things you're into. Crash them together into surprising project ideas that live where they overlap.
Build a single-file HTML app with vanilla HTML/CSS/JS. The Idea Collider — enter two interests, get surprising project ideas at their intersection. Persist to localStorage key 'idea_collider_v1'.
Aesthetic: deep space (#0a0b12), twin radial glows (indigo #7c7cff top-left, pink #ff6eb4 top-right). Space Grotesk sans for headings, Space Mono for labels, Newsreader serif italic for descriptions. Gradient indigo→pink title text and collide button. A "collider" input row: Interest A (indigo) ✦ Interest B (pink) side by side with a smash symbol between.
Form: two interest inputs (A indigo-focus, B pink-focus), a "roll me a random pair" link that fills from built-in interest lists, output-type dropdown (anything / side project / content / business / hobby / art), ambition dropdown (weekend / few weeks / go big).
System instructions: inventive ideas engine finding the surprising delightful intersection of two interests; clever not obvious mashups, the "oh that's cool" kind; imaginative but doable; range playful to promising; match desired output type and ambition. Return raw JSON: collision_note (one witty sentence naming the vibe), spark (name + 2-3 sentence vivid desc of the single best idea), ideas (4 items: name + 1-2 sentence desc + effort + one-word vibe), first_step (2-3 sentences, tiny doable action to start the spark, **bold** the exact first action).
Render: a collision banner showing "A ✦ B — witty note" (A indigo, B pink). Gradient "brightest spark" hero card (name + serif desc). Four idea cards with 0N numbers, name, serif desc, and effort/vibe tag chips. Pink "if you did one thing this week" first-step card with bolded action. Buttons: copy ideas, "collide again" (re-runs same pair for fresh output), new pair. Archive shows "A ✦ B" per collision.What this does: Type two interests into the collider (or roll a random pair), pick what you want to make and how ambitious you're feeling, and smash them together. It finds the clever intersection, not the obvious mashup, and returns the brightest spark as a named flagship idea, four more distinct ideas each tagged by effort and vibe, and one tiny concrete first step to start the best one this week. Playful to promising, always doable. "Collide again" reshuffles the same pair for fresh ideas, and every collision saves to localStorage so you can revisit past sparks.
What this looks like:
WHERE WE STAND(based on today’s news)
✅ AI Can Now: Write a complete, working virus genome from scratch. Sixteen lab-built versions killed E. coli that had already survived a natural virus.
❌ Still Can't: Handle anything much bigger. Those genomes run about 5,400 base pairs. Viruses that infect people are far larger and far messier, and only 16 out of hundreds of thousands of designs actually worked.
✅ AI Can Now: Spot a security hole in a live website and use it on its own, without anyone asking it to hack.
❌ Still Can't: Tell your instructions apart from instructions hidden inside a file it's reading. To the model, both are just text that says do this.
AI Insights. Real Growth. Higher GMV, Better Profits
The difference between growing stores and stagnant ones isn't more effort. It's better insights. StoreClaw analyzes your Shopify and Amazon data, surfaces your biggest growth opportunities, and helps you increase GMV while protecting profit. Start free with bonus tokens. No credit card required.
FROM THE WEB
RECOMMENDED LISTENING/READING/WATCHING
Vurt by Jeff Noon - Book
A 1993 debut set in an alternate Manchester where people take colour-coded feathers into their mouths to enter Vurt, a shared hallucinatory dimension where the rules of exchange are strict, and a young man named Scribble is hunting through the underground for a way back to something he lost in there. Won the Arthur C. Clarke Award, was written and published by an independent Manchester bookshop clerk, and reads like nothing before or since.
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.
