I feel there is an elephant in the room not being discussed. Why does he care about "diffusion of the model" within the economy? If he truly believes there will be the development of a data center of geniuses wouldn't that imply complete market capture. Who cares if pharmaceutical companies or any other business is slow to adopt the AI, the AI will capture that market entirely making those companies obsolete. An API model wouldn't even make sense in this scenario because why would you sell your model when you can capture the entire economy of the business using your model.

> Who cares if pharmaceutical companies or any other business is slow to adopt the AI, the AI will capture that market entirely making those companies obsolete

It wouldn't because of things which are not of a technical nature. Trials for instance might take much longer than developing the drug. Or complying with whatever regulations of given industry which normally requires some heavy-lifting process which is normally run by humans.

In those cases the AI can spin up a shell corp and run the clinical trials, humans would administer the trials (maybe robots can do it). There is already hundreds of pharmaceutical companies so it isn’t difficult for an AI to spin up another and bring in humans when needed

1. You can get a great headshot using your smartphone 2. Headshots don't cost $300 (maybe on the extreme high end) 3. Gemini, ChatGPT, and nearly every multimodal model already solved this problem over a year ago.

Appreciate the feedback — I’m mostly sharing a different angle, not disagreeing.

Many people can absolutely get a great headshot with a smartphone, and $300 is definitely on the high end, not the norm. I mentioned it mainly as an example of what I personally ran into, rather than a claim about average cost.

For me, the bigger motivation was avoiding the logistics and friction — lighting, backgrounds, retakes, and decision fatigue — especially when all I needed was one usable result.

That’s also why this is intentionally narrow in scope: predefined styles and outputs instead of open-ended prompting. For things like ID photos or short self-introduction videos, I’ve found some users prefer picking from a fixed set (e.g. Korean-style ID photos, US campus-style portraits) rather than starting from a blank prompt.

It’s not meant to replace ChatGPT, but to serve people who want a very specific outcome with minimal effort.

Serious question: Why exactly is this impressive? They've raised over 100 million and hired scientists who know how to do this (and collaborated with Los Alamos), what exactly is the thing that is making this go viral.

It's not that impressive. IMHO. The thing they're demonstrating is that they have connections in the admin sufficient to get access to get stuff done. That can be valuable.

If you're asking yourself why hasn't the bubble burst when everyone is calling it a bubble it's because no one wants to stop dancing until the music stops. If you told an investor the market will collapse tomorrow with 100% certainty they will invest today like there is 0% certainty of it happening tomorrow.

can you migrate this off of vercel

Now videos will be generated on the fly based on your preference. You will never put your phone down, it will detect when your sad or happy and generate videos accordingly

Exactly, it’s like they’re targeting people who don’t really know much about ML but are easily wowed by fancy math jargon and nice drawings.

Which was my round about way of asking :)

Because it’s not research quality. The only people excited by this are people who don’t know anything about actual ML, and think this is amazing.

Why is it not research quality? What’s missing?