What's missing?
-
What's missing? The UI is missing.
Look back at the iPhone, the iPhone didn't invent anything new. The innovation was an excellent UI that brought existing technologies together (cellphone, internet, camera) to the extent that Apple continues to be important it's due to their "better than the alternative" UI. (this has become questionable)
Part of why ChatGPT is popular is it's easy to use. I wouldn't call the UI "excellent" but it's not bad.
1/
-
What's missing? The UI is missing.
Look back at the iPhone, the iPhone didn't invent anything new. The innovation was an excellent UI that brought existing technologies together (cellphone, internet, camera) to the extent that Apple continues to be important it's due to their "better than the alternative" UI. (this has become questionable)
Part of why ChatGPT is popular is it's easy to use. I wouldn't call the UI "excellent" but it's not bad.
1/
So much of the press is focused on chips, models, technical benchmarks and not enough is focused on the quality of the user interface.
I think UI gets ignored since one of the most popular strategies to make your software company more profitable is to degrade the user experience by making it worse.
To suggest centering UI is practically alien.
2/2
-
So much of the press is focused on chips, models, technical benchmarks and not enough is focused on the quality of the user interface.
I think UI gets ignored since one of the most popular strategies to make your software company more profitable is to degrade the user experience by making it worse.
To suggest centering UI is practically alien.
2/2
@futurebird
The thing is, truly flexible multilingual natural language input *is* a great use case for large language models. It's just that that's ideally the boundary, where it converts to well-defined structured input for a testable, repeatable API of the program that does the thing. Then *maybe* they can do text generation back from the structured, well-defined output of the program.The dumb idea is making the middle bit out of floppy ,unrepeatable, untestable token generation too.
-
F myrmepropagandist shared this topic
-
@futurebird
The thing is, truly flexible multilingual natural language input *is* a great use case for large language models. It's just that that's ideally the boundary, where it converts to well-defined structured input for a testable, repeatable API of the program that does the thing. Then *maybe* they can do text generation back from the structured, well-defined output of the program.The dumb idea is making the middle bit out of floppy ,unrepeatable, untestable token generation too.
@petealexharris @futurebird agree, although I'd caveat this with the note that the current behavior of AI companies makes their versions *not worth it* even for this use case, and a second caveat that the biases inherent in the LLM concept would make me hesitate even were my first objection addressed. Like, am I sure I want to normalize UIs that require everyone to speak to computers in slightly-academic-upper-middle-class-American-English for good results? No...
-
@petealexharris @futurebird agree, although I'd caveat this with the note that the current behavior of AI companies makes their versions *not worth it* even for this use case, and a second caveat that the biases inherent in the LLM concept would make me hesitate even were my first objection addressed. Like, am I sure I want to normalize UIs that require everyone to speak to computers in slightly-academic-upper-middle-class-American-English for good results? No...
@tiotasram @futurebird
For sure. It's an accessibility accommodation, it's by no means the best way to interact with most things you might want to do with a computer. Also you don't need huge centralised models controlled by world-eating oligarchs to do simple and useful things, or want the gross homogenisation that way of doing it leads to. -
@tiotasram @futurebird
For sure. It's an accessibility accommodation, it's by no means the best way to interact with most things you might want to do with a computer. Also you don't need huge centralised models controlled by world-eating oligarchs to do simple and useful things, or want the gross homogenisation that way of doing it leads to.I think we forget that the real way people use tech is often very simple. Doing a simple task well, making it easy is the bigger revolution. If we call LLMs "the new spellcheck and translating software" that doesn't sound big enough for all the investments ... but. Having really good functional software that can do these things and be seamless is worthwhile and one could even make a little money implementing it.
But noooooo no the computer must be the new God.
-
So much of the press is focused on chips, models, technical benchmarks and not enough is focused on the quality of the user interface.
I think UI gets ignored since one of the most popular strategies to make your software company more profitable is to degrade the user experience by making it worse.
To suggest centering UI is practically alien.
2/2
@futurebird chatGPT is to LLMs what kitchen hotpads were to asbestos.
