[pen2l]

It's really hard to believe, but OpenAI kind of did Apple a bamboozle.

I use advanced voice mode in ChatGPT, and I had a bit of a eureka moment. It is the first time ever that an AI chat /actually works/. I've tried Siri, Alexa etc extensively for years, but AV mode in ChatGPT is the first AI voice chat thing that /actually works/: I can comfortably interrupt it, I can misspeak, and I can rely on it to have continuity in the context of a conversation... and the responses I get actually answer my question, most of the times. It all just works...

But for whatever reason, whatever OpenAI is giving to Apple in this Apple-OpenAI partnership, is more or less worthless. They're not giving them the keys. I wish AI voice chat would be relegated or commoditized to being just a feature (and I hope/think recent OSS advancements make that be the case), but until then I have 'Action button' set to start voice conversation with ChatGPT.

Curious to see Apple's next move though, with the recent changes in this landscape.

[jchw]

I happen to have Gemini Pro or whatever, because it's somehow bundled or free with some other Google thing I have (I don't care to ask.) Although it is a bit weird talking to a computer as if it's actually a person, I did try Google's equivalent feature in Gemini a couple of times and it seemed to work extremely well. I'm not sure how exactly it compares to OpenAI but it holds a natural conversation very well in my experience. I reckon this bodes well for Apple. It seems to me their OpenAI deal is mostly an admission that they couldn't build out their own technologies fast enough, but they certainly don't lack the capital or ambition to do it if it truly is going to be important for them in the future.

[sunaookami]

Google successfully fooled everyone into thinking that "Gemini Live" is the same as Advanced Voice Mode in ChatGPT. It's not, Gemini Live is a "stupid" speech-to-text and text-to-speech and not multimodal like AVM.

[jchw]

OpenAI's approach is certainly more technically interesting, and probably the way to go in the longer term, if the juice is worth the squeeze with this sort of technology. (After being relatively unmoved by LLMs for many other tasks, I found the voice assistant concept a lot more interesting, personally, even though I still don't have any routine uses for it.) That said, it doesn't really matter exactly how it works internally: Gemini Live accomplishes what it sets out to do, in that it feels very natural and works fairly well. I think it's clear there will be benefits to the multimodal approach of running voice directly in and out of a model for this sort of application, but if stringing together other existing technology can get you 80% of the way, it's not really a rush to get there. I don't really find this too surprising, since as far as speech recognition and voice synthesis goes, the state of the art today is very good, and most of the time computer voice interactions were greatly held back mostly by other things.

[arresin]

Is this different to the free one on ai.google? Because that I’ve tested that extensively and it is absolutely garbage in every single respect.

[ninkendo]

I’ve said this before a zillion times, but Siri is only good for imperative stuff. Commands. “Play this song”, “make a note”, “set this alarm”, etc.

And even that’s only barely true: it gets things wrong even within that very narrow set of use cases. But it at least kinda works most of the time.

Siri is absolutely not useful, and never has been, and likely never will be, when it comes to “conversational” use cases, like asking it questions or getting advice, etc.

The thing is, I only ever use voice for imperative stuff in the first place. If I want to know things or do research or have a conversation, etc, I’d much rather type into a real keyboard and read the results at my leisure. Or if I’m at a phone I can use voice to text to make it easier to do this, but it’s not really the same thing as a “conversation”.

So for me, I’m keeping Siri for the use cases it works for (home automation, timers, music, etc), because I really don’t think OpenAI will ever be good/useful for that sort of thing, even if Apple opened up the API’s to let it do so. An LLM is just too heavy-weight.

[Mistletoe]

Siri can set a timer or reminder for me sometimes. The rest is utter trash and it’s embarrassing that the largest company in the world by value, and sitting on ungodly amounts of cash, won’t or can’t improve Siri. The usual set of events is I ask Siri something and her response is so non-sensical and bad that my girlfriend and I just laugh at it and shake our heads. Then I’m reminded again why I never ask Siri anything.

[stefan_]

It's not good for the imperative stuff either, half the time some wonky popup comes to the screen. Remind me why I'm using the voice again?

[talldayo]

Looking at Apple's history in the field, it kinda makes perfect sense. They were not prepared for the seismic shift that AI was heading towards, and the hype for LLMs proved how wasted Apple's efforts were on NPU hardware. Their GPU designs are laser-focused on raster performance instead of GPGPU compute pipelines, and projects like OpenCL were abandoned under an assumption that Nvidia wasn't a real competitor.

Altman knew this was his PowerPC transition moment. He could come in there with a power-hungry, fast and attractive product that would lure Apple into a big investment to stay competitive. Apple's executives know they're beat as much as any of the engineers do, so they were probably eager to close any form of deal and reassure investors that the hype train was still very much on-rails.

[layer8]

Apple isn’t paying OpenAI anything for the ChatGPT integration: https://www.bloomberg.com/news/articles/2024-06-12/apple-to-...

[kgwgk]

> lure Apple into a big investment to stay competitive

Not so big if as they say the deal itself came at no cost to Apple.

[esskay]

Think Apple are well aware they got a bit of a raw deal with OpenAI but it was the only option on the table. Their weekness and pure stupidity with Siri became a massive elephant in the room the second chatgpt was released, and then once people saw the advanced voice mode it just made it all that more obvious how crap Apple were at AI assistants.

I dare say in 5 years time they'll be up there but for now they're heavily reliant on OpenAI to provide any guise of them having anything to offer.

[ceejayoz]

Seems risky… people whose first exposure is via iPhone seem unlikely to convert to direct users if it’s useless there.

[dingnuts]

this is still how I am with pretty much all genAI. People on this site say it's great, but a lot of people on this site have big incentives to hold that view. I have found the free models to be useless, and sufficiently useless that the paid models -- which are not cheap -- are not enticing.

Maybe I'd be more willing to pay per token if I could get a refund for all the tokens it outputs that are wrong?

Maybe my opinion of genAI will change if my employer ever allows us to use it and I am able to use a state of the art model on someone else's dime, but I'm not risking $60 or whatever to find out the performance is only moderately better than GPT3

[jasonjmcghee]

claude.ai is free (there are limits per day per account). It's not a "free [tier] model", it's their best model and (on most tasks) imo the best all around llm. The worst aspect I'd say is how outdated its information is becoming. Anything post 2023 you likely need to provide source material.

[thejazzman]

using openai or anthropic to my hearts content, configured via api to my editor, i didn't go over $10/mo -- vs $20/mo for their regular product that hit me with rate limits

[awestroke]

Yeah Kudos to Altman for pearl clutching and holding back the whole field