Apple's AI-powered Siri assistant could land as soon as WWDC 2024 | Hacker News Reader

Apple's AI-powered Siri assistant could land as soon as WWDC 2024

techradar.com

23 pointsCharlesW16 comments

For the love of God, please. Siri is so inadequate, especially compared to virtually every other voice assistant it’s incredible. I rely very heavily on ChatGPT’s voice assistant, especially in the last week or so and all I wish is that I could plug into my iPhone for basic commands

While it can’t do anything on the device, you can activate Siri and then say “Ask ChatGPT” if you have the app install and then have ChatGPT answer questions.

Is it possible in Android too?

I give it 2 weeks after release when someone prompt engineers AI Siri to say something inflammatory and the internet will explode in schadenfreude.

You can do that with Siri today. Just tell it to call someone $ADerogatoryTerm

I’m interested to see this happen but “as soon as” approximately June 2024 doesn’t feel that soon. It’s about as far from today as today is from the launch of GPT-4. Think of everything that’s happened since then. And things only seem to be speeding up. Still, I welcome a Siri that’s useful beyond setting timers.

I might actually consider the 16 Pro if Siri gets close to ChatGPT level intelligence on-device somehow. It desperately needs all the attention it can get.

I have a fear SiriGPT will get a fancy hype-filled launch, and then follow the exact same path of decay that today's Siri has followed.

IIRC, Siri's so far behind the curve that there's almost no possible way for it to catch up to more advanced assistants.

That said, China could theoretically tie Siri into its AI which could help level the playing field for both Apple and China. No idea how Apple could sell that to the American consumer.

Disappointing it wouldn't be available on the 14 pro or 15 pro, both of which have hardware that far exceeds their normal use cases. I'd be happy if it could still go to the network for some of the processing.

Processing is adequate but RAM is not. The 15 Max Pro only has 8GB RAM. You need 32 GB to run any decent sized LLMs. (You can quantize a smaller model but the quality of the output will be noticeably bad)

I have an M2 Pro with 16gb ram and even I can’t run any decent sized LLMs locally.

I would bet against Apple shipping any mobile phone with 32GB of RAM - even if the cost were acceptable the increased power draw likely would not be.

They would probably ship a specialized model trained provide specific answers on device and offload to remote for more general answers. Probably not a full Llama or Falcon.

Their latest keyboard autocomplete for instance is a GPT-2 model.

Wasn’t SIRI extremely capable at first but then Apple handicapped it? I remember something along the lines of an aquihire and high expectations but it shipped stunted. Maybe I have false memory.

Making (nearly) every AI chatbot startup obsolete overnight

Welcome to the club

Sherlocked - https://en.wiktionary.org/wiki/Sherlock#English

> 3. (computing slang) To obsolete a unique feature in third-party software by introducing a similar or identical feature to the OS or a first-party program/app. [from 2002]

Reply on news.ycombinator.com