The world is being quietly rearranged by people who write very long documents.


The title they went with Quick on the Uptake: Eliciting Implicit Intents from Human Demonstrations for Personalized Mobile-Use Agents Noisy translates that to

AI phone assistants now learn your habits — and copy them instead of just your steps


Researchers built a dataset and framework that lets AI phone assistants learn implicit preferences from watching humans, not just the explicit task steps. This means AI can now personalize its behavior to match how you actually work, not just what you're asking it to do.
Until now, AI assistants that automate phone tasks learned only the mechanical steps — tap this, swipe that. They missed the shortcuts you take, the apps you prefer, the way you do things that's distinctly you. This work shows that AI can infer those personal habits from watching you work, which changes what's possible: an assistant that doesn't just execute your request, but executes it the way you would. The catch is that this only works if the AI actually understands what you're doing — the paper's own results show a 6–7 percent improvement in alignment, which means it still gets your preferences wrong a third of the time.
Watch whether real phone assistants (Google, Apple, Samsung) incorporate this habit-learning approach in the next generation of mobile automation — if they do, we'll see whether users actually prefer personalized automation or find it creepy.

If you insist
Read the original →