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
Zuckerberg's $500M Plan to Cure All Disease With AI Cells
TLDR: Mark Zuckerberg and Priscilla Chan are putting $500 million into a project that uses AI to build virtual human cells, hoping to speed up cures for every disease.
The Story: Biohub, the nonprofit run by Mark Zuckerberg and his wife Priscilla Chan, just announced a $500 million Virtual Biology Initiative. Of that, $400 million goes to Biohub's own AI research and the rest goes to outside scientists. The five-year plan is to build AI models that can predict how human cells behave when they're healthy and when they're sick. Biohub's head of science, Alex Rives, said current datasets cover about a billion cells and the team wants ten times more than that, or even more. The data will be open and free for any researcher to use. Biohub also bought EvolutionaryScale, an AI lab that builds models for biology, and folded it in.
Its Significance: If the AI models work, scientists could test diseases and drugs on a computer instead of in a lab, which is faster and cheaper. Biohub says this could speed up cures and personalize treatments for things like cancer, diabetes, and rare diseases. Some scientists are not sure it will work. Veteran immunologist Jeffrey Bluestone told Science magazine, "It ain't going to work," though he added the project could still help in other ways. The bet only pays off if the team can collect enough cell data, and nobody knows yet how much is enough.
QUICK TAKES
The story: The Ocean's Eleven director used generative AI, in partnership with Meta, to make surreal images for parts of his upcoming Lennon doc where the audio gets philosophical and there's no archival footage to match. The film premieres at Cannes this month and was made with full support from the Lennon estate.
Your takeaway: This is the first big-name director openly using AI in a documentary about a real person, and he's doing it with permission. It's a real-world test for what counts as okay AI use in films about actual people, especially compared to the deepfake panic everyone worries about.
The story: A developer named aattaran built DeepClaude, a small script that swaps Claude Code's expensive Anthropic backend for DeepSeek V4 Pro. Anthropic's Max plan costs $200 a month, while DeepSeek costs $0.87 per million output tokens through OpenRouter, about 17 times cheaper for the same coding agent.
Your takeaway: Coding tools are getting commoditized fast. If you're spending $200 a month on Claude Code and you're not picky about which model thinks behind the scenes, this could cut your bill drastically. The trade-offs: no image input, no parallel tool calls, and Claude Opus is still better on hard reasoning.
The story: Coinbase is laying off about 700 people, roughly 14% of the company. CEO Brian Armstrong told staff the cuts are because of the crypto downturn and because AI now lets smaller teams do the work bigger teams used to. Armstrong has set a goal of 50% AI-written code at Coinbase and has fired engineers who refused to use Copilot or Cursor.
Your takeaway: This is the latest in a string of AI-blamed layoffs after Block, Pinterest, CrowdStrike, and Chegg. If your job is in tech and your company hasn't talked about AI tools yet, that conversation is coming. The likely read: AI isn't the only reason for these cuts, but it's now the easy public reason to give.
TOOLS ON OUR RADAR
🧹 BleachBit Free and Open Source: A fantastic system utility that helps you easily clean up your computer by securely deleting unnecessary files and clearing your digital footprints to free up valuable storage space.
📸 ShareX Free and Open Source: An incredibly comprehensive tool for capturing your screen recording videos and sharing your files with hundreds of different helpful features and editing capabilities.
📺 Kodi Free and Open Source: A massive media center application that transforms any computer into a beautiful entertainment hub for organizing and playing all of your movies music and television shows.
🍿 VLC Free and Open Source: A legendary and incredibly versatile media player that can play almost any video or audio file format you throw at it without needing extra codecs. The developer has turned down millions of dollars to keep advertisers from including ads.
TRENDING
Unity launches Unity AI in open beta, lets developers build games from text prompts - The game engine maker rolled out an in-editor AI agent that creates assets, scenes, and game mechanics from text. A demo trailer showed someone building a working demolition derby game in seconds. Available now to all Unity 6 developers.
Someone open-sourced a guess at what's inside Anthropic's most secretive AI model - Developer Kye Gomez published OpenMythos on GitHub, a from-scratch reconstruction of the architecture behind Claude Mythos, which Anthropic locked away in Project Glasswing after it found 271 Firefox vulnerabilities. The repo has over 10,000 stars and zero trained weights, just code and theory.
The U.S. government says China's best AI is 8 months behind. Critics say the test was rigged - NIST's CAISI evaluated DeepSeek V4 Pro using private benchmarks and a cost comparison that filtered out almost every U.S. model. Stanford's 2026 AI Index, run independently, says the gap is just 2.7%.
Colorado is rewriting its AI law after Elon Musk's xAI sued the state - Senate Bill 189 would scrap the original 2024 law and replace it with rules that require companies to tell consumers when AI made a decision about their job, housing, or loan, and let them appeal. The new start date is January 2027 instead of June 2026.
Anthropic and OpenAI both launched enterprise AI joint ventures with Wall Street on the same day - Anthropic teamed up with Blackstone, Goldman Sachs, and Hellman & Friedman in a $1.5 billion venture. Hours earlier, OpenAI raised $4 billion at a $10 billion valuation for The Deployment Company with TPG, Brookfield, Advent, and Bain. Both will embed engineers inside mid-sized client companies, copying Palantir's model.
Roomba's creator unveiled a plush robot pet that uses AI to follow you around - Colin Angle, who founded iRobot in 1990, debuted Familiar at the Wall Street Journal's Future of Everything conference. It's bulldog-sized with touch-sensitive fake fur, makes animal sounds (no talking), and is aimed at older people who don't want the responsibility of a real pet.
TRY THIS PROMPT (copy and paste into Claude, ChatGPT, or Gemini)
🧹 Dump your entire mental to-do list. Get back only what actually matters this week — and what's just anxiety dressed as productivity.
Build a single-file HTML app using vanilla HTML, CSS, JS, and one API call. Create The Noise Filter — a tool that sorts mental task chaos into signal vs. noise. Use localStorage key 'noise_filter_v1'.
Aesthetic: warm off-white (#f9f8f5), ruled lines like a legal pad, amber (#e8a020) left-border accent on the form card, black typography. Instrument Serif italic for body and prompts, Epilogue for UI, Roboto Mono for labels. Clean editorial feel with subtle ruled background lines.
Form: a large textarea for the mental dump, time horizon dropdown (this week/today/this month/next 2 hours), energy level dropdown (low/medium/high/depleted), optional win condition input.
Call the API with a system prompt that classifies ruthlessly — most lists are 30-40% signal at most. Return raw JSON: signal_ratio, verdict (one honest sentence about the list's state), signal array (task/why/action), amber array, noise array, plan (concrete realistic plan for their horizon and energy), real_ask (what this list is actually asking of them underneath the tasks).
Render: a black header bar showing signal ratio in amber and verdict text, three collapsible zone sections (Signal/Amber/Noise) with color-coded dots and task cards showing task name, why classification, and action tag, a weekly plan section, an amber-left-bordered real-ask card. First zone opens by default, others collapse. Save each filter to localStorage. History at bottom shows past verdicts. Make it work in a single HTML file.What this does:
Paste everything swirling in your head — every task, obligation, nagging thought, and "I should really." The Noise Filter sorts it into three zones: Signal (do this, it actually matters), Amber (real but not now), and Noise (anxiety dressed as productivity). It gives you a signal ratio showing what percentage of your list is genuinely useful, writes a concrete realistic plan for your time horizon based on your current energy level, and ends with one honest observation about what the list is actually asking of you underneath all the tasks. Filters save to localStorage so you can track whether your signal ratio improves over time.
What this looks like:
Say user_id. Get user_id.
Wispr Flow recognizes variable names, file references, and framework syntax mid-dictation. Speak your prompt, get developer-ready text for GitHub, Jira, or your editor. No mangled syntax. Ever.
WHERE WE STAND(based on today’s news)
✅ AI Can Now: Generate visual scenes for documentary films when no archival footage exists, with director and estate approval (Soderbergh and the Lennon estate)
❌ Still Can't: Accurately predict how a single human cell will respond to a drug, which is the whole point of Biohub's $500 million bet
✅ AI Can Now: Build a working game inside Unity from a short text prompt, including assets, scenes, and basic mechanics
❌ Still Can't: Replace a real pet's warmth and unpredictability, which is why Familiar's bulldog-sized robot uses fake fur and animal sounds instead of trying to mimic an actual dog or cat
FROM THE WEB
RECOMMENDED LISTENING/READING/WATCHING
A Cambridge mathematician walks you through how algorithms and data systems already run huge chunks of modern life, from criminal sentencing to cancer diagnosis to the music you hear next, without needing a single line of code from the reader. Fry is funny and refuses to do either the doomer or the cheerleader routine, which makes this one of the rare books that leaves you feeling like you actually understand what's happening under the hood of the world.
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.
