Dyson, Astra, and Altman Show Why 'Nothing to Hide' Works

A toothbrush with a camera went viral this week, with 17M views on a repost that wasn't even the brand’s own. 

So did a founder filming his own investor update from an exercise ball, admitting that his growth numbers were a little slow. 

Meanwhile, one of AI's most credentialed founders launched a new model to instant hype, and got fact-checked on the same day. 

OpenAI released GPT-6 Astra with a video that made desktop agents look inevitable, only for the internet to share its darker version of the story.

This week, four completely different launches got all kinds of attention. 

Let’s break them down. 

Fei-Fei Li's company, World Labs, put out a new model called Atlas on September 1st. The model generates up to a minute of camera-controlled video and reconstructs it in 3D from a handful of photographs, a truly scroll-stopping moment. The demo reel they released was genuinely impressive. 

Li, who is also known as AI's 'godmother' and creator of ImageNet, posted the second version of the announcement just after the official one went out. And within hours, early users were sharing their own clips of old photos turned into walkable scenes, creating FOMO online.

The marketing mechanism: a credible founder paired with footage that makes people believe in the impossible is a timeless combination for a product launch that got them 10M views. 

The catch:

Implicator.ai, an independent AI publication that checks industry claims against primary sources, found anomalies in Atlas’s claims. The publication didn’t find any paper, arXiv entry, model card, or code accompanying the launch.

One of Atlas's rival models had a note from its own creators, posted two weeks before Atlas launched, warning that its own scores might be inflated. World Labs used that same model as its comparison point anyway, and didn’t mention the warning. 

Atlas might be as good as the demo suggests, but because of these questions, their narrative shifted from “look what they built” to “prove it.”

OpenAI launched GPT-6 Astra on September 3rd with a single clean line: “Anything you can do on a computer, Astra can do for you. Fast.” which racked up a whopping 111.8M views in under 24 hours (and counting). The video showed the model operating a desktop end-to-end and framed Astra as the most intelligent and aligned model yet, with state-of-the-art claims across computer use, software engineering, cybersecurity, science, and professional work.

The marketing mechanism: one repeatable sentence carrying the entire announcement, with the video as proof and every qualification set aside in a table. 

The catch: 

OpenAI led with pure capability and alignment, leaving two glaring gaps for the internet to fill on its own. While the marketing prose highlighted sweeping wins, OpenAI’s own published data tables subtly revealed areas where Astra lagged behind competitors, burying the unvarnished numbers in the fine print rather than putting them in the spotlight. On top of that, the model was gated to a limited set of organisations on day one, with broader Plus, Pro, and API access promised later.

The video itself went heavily viral and pulled tens of millions of views, but the loudest reactions were not technical celebrations. High-engagement quote tweets clustered around “we had a good run,” “it’s over,” and “doomed,” with Jarvis comparisons sliding into Wall-E warnings. 

The name itself stayed Western in the official posts, Latin for the stars. Under the same announcement tweet, replies quickly noted that in Hindi and Sanskrit, Astra means a weapon. One user wrote, “Astra literally means ‘Weapon’ in the Hindi language.” 

Another said, “This is truly an Astra. If you know Sanskrit, you know I’m talking about”

The access restrictions were never explained, so the internet asked. The weaker numbers were published and never mentioned, so the internet found them. Disclosing something in a table you don't talk about buys none of the credit for candour and none of the protection from being caught. So, the conversation was now about who gets to use it, and what happens to knowledge work when the computer runs itself.

In the same week, Dyson launched a $499 toothbrush with a camera built into the brush head. According to Dyson, it has a 100,000-pixel lens that watches your teeth 28 times a second, while a model trained on 470,000 dental images fires a precision water jet into any gap it spots. And all this happens within 100 milliseconds. 

Dyson makes a bold claim of removing up to 69% more plaque than the best electric toothbrushes on the market. Their launch video is an easy ‘how-to’ explainer showing glimpses of the model working with tight macro camera angles and catchy background music.

The marketing mechanism: a bold claim, backed by precise, traceable numbers.

Dyson released its CameraJet toothbrush with full documentation. TechRadar reviewed it with a 4-star rating after using it for a week.

The rest of the job was done for free by the people on the internet. A post from Vaibhav Sisinty, unaffiliated with Dyson, pulled 17M views within 4 days of launch. He broke the pitch into four short lines, ending on “only Dyson would look at a toothbrush and think it needs a camera, a machine learning model, and a jet engine.”

And when users like @SMB_Attorney say, “$499. But you can’t put a price on health”, it created more curiosity and engagement.

Farza, founder of the YC-backed startup heyclicky, posted a nine-week update on September 1^st, sitting on an exercise ball ‘Nakey Jakey’ style. This was not the usual founder update.

The marketing mechanism: In the video, Farza tackles a question most founders are not comfortable putting out in the world: is his product working, or did it just go viral and then fade out?

Beyond the playful visuals of him bouncing on the ball and the over-the-top background images, he also shares some numbers. Some don't reflect well for his product, while others help him define success.

He mentions fifteen million views, twenty thousand downloads, a weekly growth of 10%, and gives you some math to think about. He acknowledges that while the first two numbers look great on a reel, they don't prove much on their own. He also owns the 10% growth as slow but highlights that paying customers rose from 200 to 450 in three weeks. His latest update says heyclicky has crossed 1000 paying customers.

The video works in his favor because he is naming the weak number himself. So, when he states the good numbers, they automatically lend credibility.

TIME ran Sam Altman’s interview “Inside OpenAI's Reboot” on August 26^th. This is the piece where you can see Altman introspecting in public. You’ll find him saying, “We clearly had some missteps as a company” and “we fell behind where we wanted to be.” In the article, he also admits that OpenAI became so distracted by ChatGPT’s success with consumers that its coding lead slipped.

The marketing mechanism: It’s the same as Farza, where being raw and real endears both founders to the crowd. The interview is gaining traction partly because it's Sam Altman. But Farza said almost the same thing about his own company, to a fraction of the audience, and that hit a chord just as hard. 

It’d be remiss of me if I didn’t mention Anthropic doing a similar version a few months ago. In June, researchers noticed Anthropic had capped what Fable 5 could do, without disclosing it. The backlash forced them to reverse it and they promised they'd flag it going forward. Fable 5.1 shipped a few weeks later with watermarking and content credentials as the headline feature. 

These kinds of narratives, where a brand owns its mistake and comes back stronger because of it, are a classic growth as well as marketing mechanic that work phenomenally. 

This week’s launches had what I call the “Nothing to Hide” mechanic. But it only works if you, as a brand, are doing the revealing yourself. 

A fact, a weak metric, a missing citation, a bad tradeoff, all read differently when someone else points it out. That’s why when implicator.ai found the gaps in Atlas's benchmarks, it shifted the narrative to ‘hiding something’. 

The same pattern played out with GPT-6 Astra. OpenAI led with the pure magic of a computer that can run itself. They never explained the gated rollout, and they buried the less-flattering numbers in tables nobody was pointed to.

When Farza posted his own 10% growth number, or Altman told TIME he'd been too ambitious, the bad news became proof of their transparency. So even if World Labs did put a real number out there, the defining factor came down to who got there first.

That's what makes Anthropic's move the neatest trick up its sleeve. The June backlash over Fable 5 was already public and theirs to own, so shipping the fix as Fable 5.1's headline feature closed the loop, equivalent to them saying sorry again.

Show, don't tell still holds. But there's a harder rule underneath it: tell it first. Once someone else tells it for you, the story is beyond your control. And that version will always cost more.