It's important to note that most Chinese people hardly ever use email, the Internet started rather late in China, for most Chinese people the Internet means smartphone and WeChat, and WeChat often is the only place where business communication happens, had they put a WeChat QR code on the business card, the result would be much more interesting.
Important cultural difference that the study failed to highlight.
You'd probably see a similar <15% success rate if you used a WeChat QR code for contact in the West. People might scan the code, but stop once they realize it's an unfamiliar platform that they don't have an account for.
It's a fair point, but we've tried to test this issue in a number of different ways. Based on the available data, cross country differences in email usage doesn't seem to have a meaningful impact on our results. See my comment above to gyf304.
Deepmind is google, and google suffers from chronical dabbling and never shipping products, I'd be surprised if they even care much about generalizing to useful tasks.
And about floor vs ceiling, what's really important is robustness, only robust robots can be deployed in the wild, at this time Dactyl with all its fingers is still too difficult to control, RoboCat got the grippers right, the problem really is they are again doing cute things with large models instead of raising robustness.
It can be really hard to leverage this combination of skills, the demand for cross-discipline talents is usually much lower than for a single skill set, but also much deeper when the demand exists, it's going to take some serious searching to find a match but likely it's worth it.
There was HN post about working with radiologists when building a AI diagnosis company:
Your boss could be worrying about politics in the company, asking for a raise directly can lead to that, you can find some examples in Ben Horowitz's "The Hard Thing About Hard Things", the part "HOW TO MINIMIZE POLITICS IN YOUR COMPANY".
I think here we can assume it's not someone completely ignorant in the field(they did mention "outsiders in analogous fields"), so it's really about someone with a good understanding of an adjacent field, whose skills and knowledge can often be applied to the field in question, like bringing a computer scientist to robotics or biology, not like asking an accountant to design a rocket.
All fields share some amount of border. The amount varies, but there's no such binary thing as adjacent or not, and the important insight you need may come from absolutely anywhere.
On youtube there are timer and stopwatch videos that have millions of views, people are streaming 1080p videos for something that can be implemented locally within 20 lines of code, but does it matter really, it won't make a dent on Google's revenue.
If LLMs are deployed in large enough scale, the convenience really could justify the cost.
If anything it's more like the "Early PC moment", it makes your life a whole lot easier if you use it at work and what you work on can be automated, but for most people it's more of a fun novelty, a nice to have instead of something they can't live without.
So if a parallel has to be drawn, I'll say it's WordStar and Visicalc, even the use case is similar, mostly word processing and analysis.
By delta I really mean the kind of finetuning people are doing to avoid directly giving Llama weights.
May be watermarking will be the norm even for open source models to prevent abuse.
For the government, direct technology transfer is much less of an incentive than building a domestic supply chain, which helps every other player in the industry, plus Tesla and other automakers are big taxpayers.
