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!
Mathematicians just used Meta's regular AI chat app to help answer five open questions in math. It's a good look at where these tools are useful right now: the AI drafts and calculates, and people check every step before anything counts. I'll cover what it did, what it didn't show, and a free app you can build today to check any AI answer yourself. As always, every story here is human picked and edited.
THE FRONT PAGE
Mathematicians Solved 5 Open Problems Using Meta's Regular AI Chat App
TLDR: Meta says mathematicians working with its Muse Spark AI co-wrote six research papers, and five of them answer questions nobody had settled before.
The Story:
Over several months, the researchers used the ordinary meta.ai chat in Thinking Mode, with no special setup. Muse Spark suggested proof strategies, ran calculations, wrote search programs and drafted technical sections. The people picked the problems, steered the work and checked every result.
The topics run from probability to string theory. In one paper, Muse Spark wrote a program for GAP, a math software tool, that found a 384-element example that breaks a group theory conjecture from 2024. Another settles a question about a type of wave equation that had been open since 2015.
Its Significance:
Meta published only the wins, so we don't know how many problems the AI got wrong or gave up on. AlphaSignal also notes that other teams proved three of these results on their own in August. Still, this is the same chat window anyone can open, not a lab-only system. You can casually work on difficult math problems the way you would text a friend now. If you use AI for real work, copy the method: let it draft and calculate, then check each claim yourself before you trust it. My YouTube review of Muse will be out this week, comparing it to the other recent releases. I also have a few invite codes(OK2I3S) remaining for those that haven't tried it yet. Enter it into the settings to claim the additional usage limits.
QUICK TAKES
The story: Free Gemini users will only get Flash-Lite, and AI Plus ($4.99 a month) loses access to Pro. AI Pro ($19.99 a month) keeps every model and gains Deep Think, which used to require Ultra.
Your takeaway: If you lean on free Gemini for harder tasks, test them on Flash-Lite before October 9 so you're not surprised. Google is also adding low, medium and high effort settings, and the higher ones use up your monthly limit faster.
The story: Anthropic added a setting called "Allow us to use your voice data," and it stays off unless you turn it on. It's separate from the setting that covers your typed chats and code.
Your takeaway: You can say yes to one and no to the other. Both live under Settings, then Privacy, and you can delete your voice data later if you change your mind, though it'll likely have been trained with by then.
The story: Insilico Medicine's lab in Abu Dhabi used its AI tools to design molecules that destroy BTK, a protein that helps some leukemias and lymphomas grow. Picking the candidates took 12 to 18 months, compared with 2.5 to 4 years the usual way.
Your takeaway: These are early lab compounds, so human trials are still years away. The gain is speed: more candidates tested for the same time and money, which raises the odds one of them works.
TOOLS ON OUR RADAR
🧠 Anki Free and Open Source: Flash cards that return right before you forget them, a free Quizlet Plus swap; the iPhone app costs extra.
📚 Thorium Reader Free and Open Source: Reads EPUB books, PDFs and audiobooks on your computer, a free swap for Adobe Digital Editions.
🖥️ RustDesk Free and Open Source: Fix a parent's computer from your own screen, a free TeamViewer swap that works right after install.
🎨 Krita Free and Open Source: Paint and sketch on your computer with pro brushes, a free swap for Photoshop and Procreate.
🖍️ Excalidraw Free and Open Source: Sketch diagrams that look hand-drawn right in your browser, no account needed, a free swap for Miro.
🤖 LM Studio Free: Chat with AI models that run offline on your own computer instead of paying for ChatGPT Plus. For bigger models, see Strata in Trending.
🎧 Krisp* Paid: Mutes background noise on calls and writes meeting notes; free 7-day trial, then plans from $8.
TRENDING
Attackers Used a Bug Found by Anthropic's Mythos Within a Day. A Horizon3 researcher used Mythos to find a critical flaw in the Rejetto HTTP File Server that lets attackers take full control, and attacks started within 24 hours of the disclosure. If you run HFS, update to version 3.2.1 or later.
Leaked Instructions Show Meta's Muse Keeps a Page on Everyone You Know. Wired reviewed instructions security researchers pulled from the app. They describe profiles of your family, friends and coworkers, from birthdays to old arguments, refreshed every hour. Muse has passed 5 million downloads.
Google Froze Its Open-Source Bug Bounty After a Flood of AI Reports. Google says most of the automated submissions weren't valid, so it stopped taking new reports on October 1. It plans an update in early 2027, and its other bounty programs stay open.
Trump Names a "Super Intelligence Force". Director of National Intelligence Jay Clayton leads it, joined by the FTC chair, the Pentagon's tech chief and the head of the federal personnel office. It reports to Trump and his chief of staff.
YouTubers With 300 Million Subscribers Ask for an AI Slowdown. More than 60 creators, including Mark Rober and Kurzgesagt, signed #TeamHuman, which the Center for AI Safety backs. It asks for tracking of the most powerful AI chips and a global watchdog like the one for nuclear energy.
A Free Tool Runs a 125-Billion-Parameter AI on a Gaming PC. Strata needs a graphics card with 12GB of memory and at least 32GB of RAM, and its maker reports up to 94 tokens (word pieces) a second on an RTX 5070. It's open source and keeps everything on your computer, but handles one request at a time.
TRY THIS PROMPT (copy and paste into Claude, Grok, ChatGPT, or Gemini)
🔎 Check My AI Paste any chatbot answer and see which of its claims hold up, with sources.
Build a single-file HTML app called Check My AI in vanilla HTML, CSS and JavaScript. No API key needed.
I paste the question I asked (optional) and an answer from any chatbot. The app calls the Anthropic Messages API with the web_search tool to split the answer into up to 8 factual claims, skip opinions, and search for a source on each one.
Each claim gets one label: supported, contradicted, or could not confirm. It must never invent sources, links or numbers; if it can't find a clear source, it says could not confirm.
Show a summary bar counting the three labels, a one-line verdict, then a card for each claim with what the sources say, one plain sentence on how I can check it myself, and real source links. Style: near-black background, violet and lime accents, Syne and Literata fonts.What this does: Paste an answer from any chatbot and the app pulls out each fact it states, searches the web for a source, and labels it supported, contradicted or unconfirmed. Every claim comes with a quick way to check it yourself. It can miss things too, so treat a green label as a good sign, not proof.
Your agents work while you sleep
Give a Skydive agent an ongoing responsibility and they’ll handle it on schedule, every time.
Have them prep your morning report, research new leads, monitor customer feedback, or keep projects moving overnight. You wake up, the work is already done.
WHERE WE STAND(based on today’s news)
✅ AI Can Now: Help mathematicians settle open problems from a regular chat window, including writing a program that found a 384-element counterexample.
❌ Still Can't: Vouch for its own proofs. People checked every result, and Meta didn't share the attempts that failed.
✅ AI Can Now: Narrow down blood cancer drug candidates in 12 to 18 months instead of 2.5 to 4 years.
❌ Still Can't: Show those drugs work in people. They're early lab compounds, years away from human trials.
RECOMMENDED LISTENING/READING/WATCHING
🎧 Lex Fridman Podcast: Terence Tao (#472) - Podcast
Terence Tao, one of the best-known mathematicians alive, spends part of this 2025 conversation on AI-assisted proofs, the Lean proof checker and whether an AI could ever win a Fields Medal. It's long, so jump to the AI section about an hour and a half in. It pairs well with today's lead because it covers how mathematicians check computer-assisted work, the part Meta's announcement doesn't show.
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.
