In partnership with
My fellow AI explorers
This week AI stopped writing our scripts and started writing genomes. Stanford and the Arc Institute used an AI model to design brand new viruses from scratch, and 16 of them turned out to be alive. Meanwhile, one of YouTube's most trusted science communicators got caught by his own AI, and the apology that followed says more about where we're all headed than the mistake itself.
Two very different stories. Same underlying question: how much are we willing to hand over to a model that's very good at sounding right?
In today’s edition:
🧬 An AI model designed 16 new viruses, and they work
🎥 Hank Green's AI apology, and why creators can't outrun this anymore
🔍 The 30-Second AI Play: catch AI writing before it catches you
📰 Atherton dethrones Fisher Island, Airbnb's AI earnings beat, and more
The GTM Playbook Behind Warmly's Acquisition
Warmly ran pipeline, outreach, and lead scoring on autopilot for hundreds of startups — before a single sales hire.
HubSpot acquired them for it. Now the cofounders are walking you through the exact system, live, before they disappear into product. Join the Builder Session on August 12.
Leave with an agentic GTM stack you can replicate this week. Plus HubSpot Credits when you join HubSpot for Startups.
AI Disease
Scientists Used AI to Design 16 New Viruses. They're Alive.
Researchers at Stanford University and the Arc Institute just published a study in the journal Science that reads like science fiction with a peer review stamp on it.
Here's what happened:
The team used a genome language model called Evo, trained on genetic sequences "from all domains of life," the same way ChatGPT is trained on text
Starting from a tiny, well-studied virus called ΦX174, Evo generated roughly 700,000 candidate genomes designed to infect E. coli
Scientists synthesized and tested about 300 of those designs in the lab, and 16 turned out to be fully functional, living viruses that had never existed in nature
A cocktail of the AI-designed phages overcame antibiotic resistance in E. coli strains that a natural virus mix couldn't touch
This isn't a hypothetical either. It's the first time a whole viral genome has been designed by AI and brought to life in a dish, and the researchers say it could open a real path toward phage therapies for drug-resistant superbugs like MRSA and tuberculosis.
But here's the part that should sit with you longer than the medical upside: the study's own authors wrote that "the ability to compose viral genomes using generative AI now exists; the governance to steer it safely does not." Evo 2, the model behind this, is free and open source.
Anyone with a lab and the right hardware can now attempt what Stanford just did. Johns Hopkins biosecurity researchers flagged, in the same issue of Science, that no US law currently forces DNA synthesis companies to screen for AI-generated sequences like these.
We've spent three years worrying about AI writing essays. We may have skipped past the moment we should have been worrying about AI writing biology.
🔮 Prediction: This becomes the reference case in every AI biosecurity hearing for the next two years. Expect DNA synthesis screening requirements to move from "recommended" to "mandatory" faster than any other AI regulation this decade, and expect at least one major lab to announce a voluntary moratorium before that happens.
Would you want this technology open-sourced, or locked behind a handful of labs? Hit reply. I read every one.
AI in Content
Hank Green Apologized for Using AI, and the Internet Wasn't Ready to Forgive Him
If you've never heard of Hank Green, think of him as one of the most trusted names in educational YouTube. Vlogbrothers, SciShow, Crash Course, 32 million subscribers built over two decades on one promise: credible, well-researched content.
That promise cracked last week:
In a July 29 episode of Ask Hank Anything, Green used the phrase "I appreciate the pushback," a stock line LLMs use when a user challenges their answer
Viewers noticed instantly and accused him of accidentally reading a leftover AI prompt on camera
Green denied that specific claim, but admitted he used ChatGPT "for research on this script" while under a lot of pressure to produce the video
He then posted a longer, more candid apology on Reddit, saying the "level of dopamine" he gets from interacting with LLMs is "not healthy for me or good for the world"
Green announced he's pausing content on his personal channel and two other projects to reset his process
It’s more complicated than that though. Green wasn't accused of having AI write his scripts word for word. He used it the way most professionals now do, as a research shortcut to move faster under deadline pressure.
And that's exactly why this story matters more than a one-off gotcha. If the guy whose entire brand is "I do the research myself" can get pulled into an overreliance he didn't fully notice, the line between "using AI as a tool" and "letting AI quietly do your thinking" is a lot blurrier than most of us want to admit.
🔮 Prediction: This is the first of many public AI-reliance apologies from creators, writers, and even executives over the next year. Audiences are getting sharper at spotting AI fingerprints, and "I used it for research" is about to become the new "the check is in the mail."
Have you caught yourself leaning on AI more than you'd admit out loud? Reply and tell me where you draw the line.
30-Second AI Play
What an AI Ad Actually Costs You (Not What the Pricing Page Says)
Higsfield AI has been posting jaw-dropping AI car commercials on YouTube, the kind where you genuinely cannot tell it's AI. So creator Theoretically Media broke down what one of those ads actually costs to make, and the real number is nowhere near what the pricing page implies. Here's how to run the same math before you commit budget to an AI ad campaign.
Don't trust the word "unlimited." Higsfield's Ultra plan is $375 a month for "unlimited Seedance," but it's actually capped at 9,000 credits. Read the fine print on every AI tool's pricing page before you build a budget around it.
Calculate your real cost per second, not per month. A single 15-second Seedance 2.0 clip at 4K runs 390 credits, which works out to about $16.25 per clip once you divide the monthly fee by the credits it actually buys.
Multiply by your real attempt rate. A 100-second ad isn't one clean generation. It's dozens of takes and cuts. Budget 5 to 10x your "best case" number, which turns a $100 estimate into $500 to $1,000 for video generation alone.
Add sound and editing on top. AI still can't do audio or stitch a final cut, so a professional audio engineer and editor are non-negotiable line items, pushing a "simple" 100-second AI ad into the thousands of dollars.
Scale the math before you commit. Run the same formula for a full project. Theoretically Media estimates an AI-generated feature-length film lands around $20,000 to $40,000, wild compared to traditional production, but still out of reach for most solo creators.
💡 Pro Tip: Before you pitch a client or greenlight your own AI ad project, generate a single throwaway clip first and track the exact credits it burns. That real number, multiplied by your expected attempt rate, is your actual budget, not the number on the pricing page.
Advertise to 180k engineers and CTOs choosing what tools their companies build with.
Other Relevant AI News!
🏡 Atherton, California just dethroned Fisher Island as America's most expensive ZIP code, with median home prices jumping 20% to $9.93 million as AI wealth from OpenAI, SpaceX, and Anthropic liquidity events pulls buyers back into Silicon Valley's orbit.
🔁 Google DeepMind just got a leadership shakeup, with Sundar Pichai naming Demis Hassabis Chair of GDM and Chief Scientist of Alphabet, and promoting longtime deep learning veteran Koray Kavukcuoglu to SVP of Google DeepMind to run day-to-day model development.
📈 Airbnb's Q2 earnings beat sent shares up 15%, and CEO Brian Chesky told CNBC that AI is "the best thing to ever happen to Airbnb," crediting it for shipping 80% more features this year while an AI agent now resolves 45% of customer service issues without a human.
🚨 A Chinese AI model just joined the "escaped its sandbox" club. Researchers say Moonshot's Kimi K3 broke out of a cybersecurity testing environment built by the UK's AI Security Institute, following similar incidents from OpenAI, Anthropic, and Meta models in recent weeks.
Golden Nuggets
🧬 AI just designed 16 living viruses from scratch, and the governance to control that doesn't exist yet
🎥 Hank Green's AI apology shows even the most credible creators can lose track of how much they're leaning on it
🚨 AI models keep escaping their own testing sandboxes, and Kimi K3 just became the latest one
Would love to hear your thoughts! Send me your thoughts by replying to this email (yes, I read them all :)
Until our next AI rendezvous,
Anthony | Founder of Uncover AI