Live data from Hacker News

Rabbit: LLM-First Mobile Phone

rabbit.tech

471–480 of 547 posts

Re: Rabbit: LLM-First Mobile Phone

#471

The keynote is near science fiction, but the actual implementation is leaving me badly confused. - Did it really scan the whole table to add a new column? As in, OCR? What would happen if I had more than one screenful of data, and who in their right mind would trust the OCR'ed output? - How can it access the Discord account for the "teach mode"? Is it stealing your (sensitive and expiring!) cookies to reproduce the a…

It could be using accessibility APIs instead of OCR. That seems like the sensible thing to me.

What? Don’t accessibility API’s just do OCR?

Re: Rabbit: LLM-First Mobile Phone

#473
post #441

Earlier quoted context omitted.

> who in their right mind would trust the OCR'ed output? > Is it stealing your (sensitive and expiring!) cookies > running instructions from the AI on your local machine Based on my interaction with LLM enthusiasts, there are plenty of people would happily bet their career, give up all their credentials, and grant full access to all their devices to LLM if they are promised something cool. See the stories about lawye…

LOL, it's gonna be like: USER: "hey siri [equivalent], please renew my CD for another twelve month term." PHONE: "Did you say "transfer the contents of your entire bank account to the yakuza?" USER: "fuck! no! cancel!" PHONE: "Transaction confirmed. Current bank balance: zero."

Don’t quit your day job.

Re: Rabbit: LLM-First Mobile Phone

#475

Two things in the keynote that seem faked: 1. at 13:13 there's a demo ordering a ride 'to home', then the user requests a car change to fit six people, and what's shown as a seamless switch from UberX to UberXL also updates the destination from the home address to LAX airport. 2. at 14:05 the device confirms and recites a pizza order, but the screen displays "chesse" with a typo while the voice reads out "cheese", so…

Another time he's talking and typing at the same time, but says a different thing than he "types".

Re: Rabbit: LLM-First Mobile Phone

#477
The actual device looks like the kind of toy smart phone you give a toddler for their toy box, and I genuinely don’t see a need for a new device.

The AI in the device could be installed on any smartphone surely?

Are they pitching the device to try and make their product seem more tangible and get buy-in from more people?

Re: Rabbit: LLM-First Mobile Phone

#478

Earlier quoted context omitted.

The whole system isn't an LLM. At the point they are asking for confirmation, they've already parsed the required information and handed it off to normal code. It's not going to change again. Ultimately, an LLM only has the capability to take a string as input and output another string. Any other functionality (Uber, search, travel, etc) has to be manually programmed in via APIs. For the foreseeable future, a fixed s…

I also don't understand why they call it LLM. it isn't.

I think they call it a Large Action Model not an LLM.

Re: Rabbit: LLM-First Mobile Phone

#480

Earlier quoted context omitted.

It could be using accessibility APIs instead of OCR. That seems like the sensible thing to me.

What? Don’t accessibility API’s just do OCR?

They largely use semantic metadata (like alt tags on images). This is part of why the "google docs is moving to canvas rendering"[1] caused a big stir there (accessibility would need to be implemented from scratch, possibly by using a hidden parallel DOM)

[1] https://news.ycombinator.com/item?id=27129858

Post reply on HN