Post ATjM4p5PtMQFZm7fwe by corb@mastodon.online
(DIR) More posts by corb@mastodon.online
(DIR) Post #ATjKn3Pxlm1yrDhFIG by thomasfuchs@hachyderm.io
2023-03-18T02:59:54Z
0 likes, 1 repeats
Just ran across the funniest shit, someone saying “Apple is so behind on AI”.Meanwhile I’m using Apple’s AI stuff every time I make a photo with my phone, every time I search for images of stuff on my phone, when copying a photo without background, when copy & pasting text from an image, etc. etc.I guess it’s true that Apple is behind on bullshit generators and lying search engines.
(DIR) Post #ATjKn3zlcefeeGRq6K by film_girl@mastodon.social
2023-03-18T05:08:04Z
0 likes, 0 repeats
@thomasfuchs I think the argument that Apple is behind on AI is definitely wrong, but I think a reasonable argument is that Apple is behind on AI tools and APIs for third-party devs. Core ML and Create ML are not as good as competitors, that’s a fact. And the gap is getting wider. And that’s to say nothing of the hindrance not being able to run CUDA poses for us Mac users.
(DIR) Post #ATjKn4Y9YoB0MuXIhM by thomasfuchs@hachyderm.io
2023-03-18T05:16:20Z
0 likes, 0 repeats
@film_girl That’s fair, my context here was about tech bros who are insinuating that AI is purely about LLMs somehow (because that’s the hype they’re currently into).
(DIR) Post #ATjKn53hfVPhwlIUsK by simon@fedi.simonwillison.net
2023-03-18T05:26:50Z
0 likes, 0 repeats
@thomasfuchs @film_girl Did you see this? Frustratingly I think there's a good chance that "AI" referring to generative models may rapidly become the unavoidable common meaning, in the same way that "crypto" doesn't mean cryptography for most people any more: https://simonwillison.net/2023/Mar/15/ai/
(DIR) Post #ATjKn66vkzabD3oaXo by thomasfuchs@hachyderm.io
2023-03-18T03:02:37Z
0 likes, 0 repeats
They’ve been designing, shipping (and using!) their own neural network hardware for years, for fuck’s sake. Likely by far the largest installed base of neural network accelerators with hundreds of millions of devices.
(DIR) Post #ATjKySRV4ILbeiqxLU by film_girl@mastodon.social
2023-03-18T05:27:48Z
0 likes, 0 repeats
@simon @thomasfuchs ugh. That is frustrating.
(DIR) Post #ATjLGJhkIK6Plfy3rE by film_girl@mastodon.social
2023-03-18T05:30:09Z
0 likes, 0 repeats
@simon @thomasfuchs I am probably annoyingly fastidious in referring to generative AI as much as possible precisely because I want to be specific with what we are talking about. The generative stuff is the most exciting for me personally right now, but I would never pretend it represents all or even most of AI, nor that it should become the new name.
(DIR) Post #ATjLGKFQH72bS7ixLk by simon@fedi.simonwillison.net
2023-03-18T05:32:19Z
0 likes, 0 repeats
@film_girl @thomasfuchs Yeah I've been fastidious about using the "generativeai" tag on my blog, but I do worry that the more general audience that I'm hoping to reach with some of my writing won't know what "generative AI" means, or what "large language models" are eitherI really don't want to have to stick the "ChatGPT" term on everything just so people know what I'm talking about though!
(DIR) Post #ATjLWZKXxwTurcrldw by film_girl@mastodon.social
2023-03-18T05:34:54Z
0 likes, 0 repeats
@simon @thomasfuchs same. Although my short-pitch/explainer for GitHub Copilot has been “ChatGPT for code” since December. The irony is Copilot is 18 months older than ChatGPT, uses the same GPT engine, has had some of the fame features (explain code block, ask it to write code for you) but has maybe 1% of the name recognition.
(DIR) Post #ATjM4p5PtMQFZm7fwe by corb@mastodon.online
2023-03-18T05:41:05Z
0 likes, 0 repeats
@simon @thomasfuchs @film_girl At my company you can’t say your project is going to use AI. It’s basically meaningless now. You have to say specifically what technology you are using.
(DIR) Post #ATjiVolTOTxNOA4VFY by kazaroth@mastodon.social
2023-03-18T09:52:17Z
0 likes, 0 repeats
@simon @film_girl @thomasfuchs I’m more in the camp that ‘AI’ is a bad name for any of this stuff (including LLMs - which is probably the best name). Unfortunately ‘AI’ ending up meaning ‘only’ the language models, as if they’re more intelligent is almost worse. I think the most dangerous rhetoric is the tech bros who keep saying: ‘well maybe the brain is just a statistical model too!’. In some sense it may be, but it’s infinitely more complex than a load of edge weighted language tokens.
(DIR) Post #ATjsIi3u0D0GNT51Pc by simon@fedi.simonwillison.net
2023-03-18T11:42:23Z
0 likes, 0 repeats
@kazaroth @film_girl @thomasfuchs I argued against that "what if the brain is just a large language model?" question on KQED radio a few weeks ago - transcript at the bottom of this post https://simonwillison.net/2023/Mar/7/kqed-forum/