5 ms·

a little cope for the fact that apple failed to present state of the art performance

by seydor 2y ago

Closi 2y ago
Apple seems to be betting that something polished which is “less than state of the art” will sell better than something less polished and more advanced. Technical capability is rarely more important than product market fit.
- sigmoid10 2y ago
  The problem here is that for current AI systems, the state of the art is the one that has ironed out the most kinks. I.e. what you would expect from a polished product. That is basically ChatGPT with it's monstrous API that does tons of god knows what behind the scenes in addition to feeding text to an LLM, most of which was figured out by trial and error from user reports. There's no doubt that a modern LLM based Siri would be a much better product than the old one, but if they're not careful, they will fall into the same trap they did the last time introducing an assistant.
  - tomashubelbauer 2y ago
    Privacy is a part of the equation and while Apple could gather all the context off my phone's local data and give ChatGPT the whole thing to really give it the best shot and providing a useful and helpful answer, this would not be a polished product on all axes, mainly the responsiveness and privacy ones. The tradeoff of using local models for less than SoTA inferrence but with unbeatable privacy and latency seems right to me for Apple and what Apple customers want and expect.
  - Closi 2y ago
    I think this is still a tech-centric view rather than a product-oriented view. E.g. Apple generating emojis using a diffusion model is clearly less SOTA than other image models, but could be a really neat feature from a product/user perspective. Would a better diffusion model help? Maybe, but it probably changes the product less than all the UI and tight integration work they will have done. If I was to guess, I would guess many more emoji's will get created through this feature than images generated in Microsoft's 'Image Creator' in Paint (I could be wrong - just my gut feel).
    - viraptor 2y ago
      WhatsApp has integrated AI generated stickers almost a year ago. (https://www.theverge.com/2023/8/15/23832635/whatsapp-ai-generated-stickers-test-android-beta https://www.theverge.com/2023/8/15/23832635/whatsapp-ai-gene...) I'm not sure how much they're used globally, but there are people crazy about stickers and the quality of what I got was really decent already. There were already external apps available for this purpose for a while, so yeah, the interest is there.
      - fennecbutt 2y ago
        Microsoft SwiftKey has also been doing this for quite a while (not quite that long) as well.
  - HarHarVeryFunny 2y ago
    The best AI experience right now isn't just using SOTA models, but using them in a way that plays to their strengths and avoids their weaknesses (hallucination, limited reasoning, etc). The way to minimize hallucination is to use LLMs to manipulate text (translate, summarize, query, reword, etc) rather than generate it out of thin air, and arguably Apple are doing this by integrating LLM use into the product and therefore controlling it's application, rather than just providing a chat interface that people are wont to use an hallucinating "search engine"/oracle.
- andruby 2y ago
  Indeed. And that has always been Apple’s forte.
  - pjmlp 2y ago
    Not always, specially when Steve Jobs was away.
  - tomalbrc 2y ago
    Except for the dozens of products that were neither polished nor state of the art (maps, Siri, iOS 6/7, most of their Mac related software, etc)
Hippocrates 2y ago
Why does SOTA matter here? OpenAI has actually failed already to make something useful and mass-market. After all these years of pushing SOTA, it's still a website with a text box. Siri in its current form is more useful to me than ChatGPT 4. You simply don't need SOTA models for a lot of valuable features; you need context, annotation APIs, and the right platform and integrations to assist users where they already are (on their phones, in cars, in their IDE, etc). MSFT's GitHub copilot is a good example of getting it right. Apple relegated ChatGPT to a 3rd tier AI capability for "google answers", and "write me a poem" party tricks, and rightfully put it behind a privacy disclaimer. It looks barely better than the various ChatGPT shortcuts so many people have cobbled together. That part of the announcement stuck out like a sore thumb and looked to me like a huge L for OpenAI. The "partnership" was a nothingburger. Like someone at Apple agreed to it early on but then late in the project realized it wasn't needed at all. So much for SOTA.
- nicce 2y ago
  > Why does SOTA matter here? OpenAI has actually failed already to make something useful and mass-market. It is mainly B2B market in reality. Look for the API capabilities: https://platform.openai.com/docs/introduction https://platform.openai.com/docs/introduction
- infecto 2y ago
  I mostly agree with you. OpenAI is reportedly still doing great in terms of revenue, but Apple's implementation is magical if it performs as shown in the keynote. In my opinion, it's the best implementation of LLM/AI in a consumer device. It's amusing to think back to all the buzz around Humane's AI pin.
  - Hippocrates 2y ago
    > It's amusing to think back to all the buzz around Humane's AI pin It certainly is. And the Rabbit R1. How in the world did people who supposedly know anything about AI think they could make that work as a standalone device detached from rich context. The sad thing is, rich context may not be possible outside of OS-level integration that is gate kept, but I still think they're idiots for trying.
threeseed 2y ago
It is best in class for a model that runs primarily on a mobile device.

a little cope for the fact that apple failed to present state of the art performance

by seydor 2y ago

a little cope for the fact that apple failed to present state of the art performance

Apple seems to be betting that something polished which is “less than state of the art” will sell better than something less polished and more advanced. Technical capability is rarely more important than product market fit.

sigmoid10 2y ago
The problem here is that for current AI systems, the state of the art is the one that has ironed out the most kinks. I.e. what you would expect from a polished product. That is basically ChatGPT with it's monstrous API that does tons of god knows what behind the scenes in addition to feeding text to an LLM, most of which was figured out by trial and error from user reports. There's no doubt that a modern LLM based Siri would be a much better product than the old one, but if they're not careful, they will fall into the same trap they did the last time introducing an assistant.
- tomashubelbauer 2y ago
  Privacy is a part of the equation and while Apple could gather all the context off my phone's local data and give ChatGPT the whole thing to really give it the best shot and providing a useful and helpful answer, this would not be a polished product on all axes, mainly the responsiveness and privacy ones. The tradeoff of using local models for less than SoTA inferrence but with unbeatable privacy and latency seems right to me for Apple and what Apple customers want and expect.
- Closi 2y ago
  I think this is still a tech-centric view rather than a product-oriented view. E.g. Apple generating emojis using a diffusion model is clearly less SOTA than other image models, but could be a really neat feature from a product/user perspective. Would a better diffusion model help? Maybe, but it probably changes the product less than all the UI and tight integration work they will have done. If I was to guess, I would guess many more emoji's will get created through this feature than images generated in Microsoft's 'Image Creator' in Paint (I could be wrong - just my gut feel).
  - viraptor 2y ago
    WhatsApp has integrated AI generated stickers almost a year ago. (https://www.theverge.com/2023/8/15/23832635/whatsapp-ai-generated-stickers-test-android-beta https://www.theverge.com/2023/8/15/23832635/whatsapp-ai-gene...) I'm not sure how much they're used globally, but there are people crazy about stickers and the quality of what I got was really decent already. There were already external apps available for this purpose for a while, so yeah, the interest is there.
    - fennecbutt 2y ago
      Microsoft SwiftKey has also been doing this for quite a while (not quite that long) as well.
- HarHarVeryFunny 2y ago
  The best AI experience right now isn't just using SOTA models, but using them in a way that plays to their strengths and avoids their weaknesses (hallucination, limited reasoning, etc). The way to minimize hallucination is to use LLMs to manipulate text (translate, summarize, query, reword, etc) rather than generate it out of thin air, and arguably Apple are doing this by integrating LLM use into the product and therefore controlling it's application, rather than just providing a chat interface that people are wont to use an hallucinating "search engine"/oracle.
andruby 2y ago
Indeed. And that has always been Apple’s forte.
- pjmlp 2y ago
  Not always, specially when Steve Jobs was away.
- tomalbrc 2y ago
  Except for the dozens of products that were neither polished nor state of the art (maps, Siri, iOS 6/7, most of their Mac related software, etc)

Hippocrates 2y ago

Why does SOTA matter here? OpenAI has actually failed already to make something useful and mass-market. After all these years of pushing SOTA, it's still a website with a text box. Siri in its current form is more useful to me than ChatGPT 4. You simply don't need SOTA models for a lot of valuable features; you need context, annotation APIs, and the right platform and integrations to assist users where they already are (on their phones, in cars, in their IDE, etc). MSFT's GitHub copilot is a good example of getting it right. Apple relegated ChatGPT to a 3rd tier AI capability for "google answers", and "write me a poem" party tricks, and rightfully put it behind a privacy disclaimer. It looks barely better than the various ChatGPT shortcuts so many people have cobbled together. That part of the announcement stuck out like a sore thumb and looked to me like a huge L for OpenAI. The "partnership" was a nothingburger. Like someone at Apple agreed to it early on but then late in the project realized it wasn't needed at all. So much for SOTA.

nicce 2y ago
> Why does SOTA matter here? OpenAI has actually failed already to make something useful and mass-market. It is mainly B2B market in reality. Look for the API capabilities: https://platform.openai.com/docs/introduction https://platform.openai.com/docs/introduction
infecto 2y ago
I mostly agree with you. OpenAI is reportedly still doing great in terms of revenue, but Apple's implementation is magical if it performs as shown in the keynote. In my opinion, it's the best implementation of LLM/AI in a consumer device. It's amusing to think back to all the buzz around Humane's AI pin.
- Hippocrates 2y ago
  > It's amusing to think back to all the buzz around Humane's AI pin It certainly is. And the Rabbit R1. How in the world did people who supposedly know anything about AI think they could make that work as a standalone device detached from rich context. The sad thing is, rich context may not be possible outside of OS-level integration that is gate kept, but I still think they're idiots for trying.

threeseed 2y ago

It is best in class for a model that runs primarily on a mobile device.