- Uncovering AI
- Posts
- 🖼️ Meta made your face an AI prompt. By default.
🖼️ Meta made your face an AI prompt. By default.
Instagram opted you in without asking. Here's how to opt back out.
In partnership with
My fellow AI explorers
I spent part of this week digging through opt-out settings on an app I didn't even remember installing, and I have a feeling a lot of you are about to do the same thing.
That's the mood of this edition. Meta shipped its first real image model and buried a genuinely uncomfortable default inside the rollout. xAI dropped Grok 4.5 with pricing and efficiency that should make every other lab nervous. And I'm handing you a copy-paste prompt that turns any model into a website design machine, no framework required.
In today’s edition:
🖼️ Meta's new Muse Image model quietly opts your Instagram photos into AI remixing
⚡ Grok 4.5 undercuts the competition on price while beating them on tokens per task
🎨 The one prompt that gets any model to design 25 websites while you do something else
How owning AI deployment expands your career
Across product, ops, and CX teams, a new kind of role is taking shape: the person responsible for making AI actually work, day to day.
On July 16, three people living this shift join a live roundtable: Simone Santiago Broad (Yoco), Yelva Espinoza (Zumba Fitness), and Fin's Dave Lynch. You'll hear what the job really looks like across industries, how they carved out these roles, the skills they'd hire for, and the challenges they're tackling now. Bring your questions, since the best moments happen live.
Register for the roundtable to save your spot.
Meta
Meta's Muse Image Launch Comes With a Consent Problem
Meta just shipped Muse Image, the first media generation model built by Meta Superintelligence Labs. On paper, it's a genuinely strong release.
The model works agentically instead of just mapping a prompt straight to pixels. It writes and runs code to nail things like charts and QR codes, searches the web to ground images in real facts, and reflects on its own drafts mid-generation, redoing a section or starting over entirely when something's off.
None of that self-correcting behavior was designed in. Meta says it just emerged during reinforcement learning because it produced a higher reward. Give it more test time to compute, and it keeps getting better, in a roughly log-linear curve, similar to what we've seen with reasoning LLMs. Right now, it holds the number two spot on the image generation leaderboard.
Here's the part that's generating actual backlash.
Instagram integration is opt-out, not opt-in. Anyone can @-mention a public Instagram account inside a Muse Image prompt and pull that person's real photos in as visual reference, no consent, no notification.
You won't be told if it happens to you. Meta's own help page confirms it plainly: you will not be notified about content created using AI features at Meta.
Opting out doesn't undo anything already made. Turning off the toggle in Sharing and reuse stops future generations, but any image someone already created using your face stays exactly where it is.
Meta frames this as a creative personalization tool, the kind of thing that helps you mock up an event invite or a collaborative concept. Fine in theory. In practice, it means every public Instagram account is now fair game as raw material for a stranger's AI prompt, and the burden is entirely on the user to go find the setting and turn it off.
🔮 Prediction: this becomes Meta's next regulatory headache before it becomes a beloved feature. The EU AI Act and the UK's Data Use and Access Bill are both pushing toward stricter consent standards for exactly this kind of use, and "we didn't notify you because our help page technically says we won't" is not going to hold up well under scrutiny.
Reply and tell me: did you know your public Instagram photos were opted in by default, and did you turn the setting off once you found out?
xAI
Grok 4.5 Is the Cheapest Smart Model on the Market Right Now
xAI dropped Grok 4.5 today, and the headline isn't just raw intelligence: it's what the intelligence costs you.
Grok 4.5 was trained alongside Cursor and built specifically for coding, agentic work, and knowledge tasks. It's serving at 80 tokens per second, which puts it in flash model speed territory. And it uses roughly 4.2 times fewer output tokens than Opus 4.8 to resolve the same SWE Bench Pro task. That efficiency gap is the real story.
Pricing that undercuts the field. $2 per million input tokens and $6 per million output tokens… It’s meaningfully cheaper than comparable frontier models.
Office work, not just code. Grok 4.5 is the new default model in Grok Build, and it's now shipping inside Word, Excel, and PowerPoint plugins, building multi-sheet financial models and full slide decks from a single prompt.
Benchmarks land mid-pack, not top of the leaderboard. On DeepSWE 1.0, it scores 62%, behind Fable's 66.1% and GPT 5.5's 64.3%, but ahead of Opus 4.8's 55.8%. The pitch isn't that it’s the "smartest model." It's "smart enough for a fraction of the cost and tokens."
One thing to flag if you're building in Europe: Grok 4.5 isn't available yet in the EU through any xAI product or the API console. xAI says that's coming in mid-July.
🔮 Prediction: this is the model that finally makes "intelligence per dollar" a headline metric instead of a footnote. When a model resolves the same task in a quarter of the tokens at a third of the price, enterprises stop asking who's smartest and start asking who's cheapest to run at scale. Expect the other labs to answer with their own efficiency numbers within the month.
Reply and tell me: are you picking models based on raw benchmark scores, or has cost per task already become your deciding factor?
30-Second AI Play
Turn One Prompt Into an Infinite Website Design Machine
Want a model to design a genuinely impressive website with zero back and forth? Skip the multi-step framework. Give it one ambitious prompt and get out of its way.
Pick your model and give it real tools. Feed it access to an image library like Pinterest for visual inspiration, an image or video generation tool, and a place to host the result.
Write one prompt that sets the bar high, not the steps. Ask for a fundamentally different, visually striking site using advanced techniques like 3D, custom animation, and unique typography, and give it total creative freedom on execution.
Tell it to host the result and document how. Ask it to publish to a service like Netlify and write a short guide explaining its own process so you or anyone else can repeat it.
Require iteration passes before it calls anything done. Ask for at least three self-review passes per site, where it goes back through looking for design problems and opportunities to improve.
Set it loose and don't check in. Tell it to work autonomously and not ask for anything until it's finished, then verify the output yourself once it's done, either visually in a desktop app or with a browser tool.
💡 Pro tip: the instinct to give a model a rigid four-step design framework is usually what holds the output back. The more creative freedom and the clearer the finished goal, the better the result. Treat verification, not micromanagement, as your actual job here.
Advertise to 180k engineers and CTOs choosing what tools their companies build with.
Other Relevant AI News!
🇨🇦 Meta is building its first major Canadian data center, part of the company's broader push to lock down AI infrastructure outside the US.
📈 Nvidia stock climbed after reports that Beijing plans to let major Chinese firms like Alibaba and ByteDance buy a limited number of H200 chips, though shipments haven't started yet.
🏛️ OpenAI is expanding GPT-5.6 access, ending the government-imposed limits that had confined the model to a small list of vetted partners.
🔋 Tesla launched Tesla Home, a consumer-facing dashboard built on its Opticaster AI engine that's already run over 100 million hours managing batteries across Tesla's fleet.
Golden Nuggets
🖼️ Meta's Muse Image launch is impressive tech wrapped around an opt-out consent problem that's going to keep making headlines.
⚡ Grok 4.5 is betting that the cheapest and most efficient beats the smartest, and the pricing backs that bet up.
🎨 The best model outputs come from ambitious prompts and real freedom, not rigid frameworks.
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