The elephant in the room: The user interface problem
We seem to dancing around a problem in the middle of the room like an elephant no one is acknowledging, and that is the interface to Artificial Intelligence and Generative AI is a place that requires several degrees of innovations.
I would argue that the first winning feat of innovation on interfacing with AI was the "CHAT BOX". And it works well enough for the 40% of use cases. And there is another 20% of uses that WE THE PEOPLE can use our imagination (prompt engineering) to manipulate the chat box to solve. On this topic, there was an article/opinion that said complex LLMs are unnecessary because 90% of people don't need it. Yeah. Because the chat box cannot do much more that would require heavier LLMs.
Complex AI and large data sets need nicer presentation and graphics, more actionable interfaces, and more refined activity concepts, as well as metadata that gives information on the reliability or usability of generated information.
Things like edit sections of an article, enhance articles, simplify articles, add relevant images, compress text to fit in a limited space, generate sql data from these reports, refine patterns found in a page with supplied examples, remove objects, add objects, etc.
Some innovation has to happen in MS Office interfaces. Some innovations have to happen in photoshop-like interfaces.
The author is complaining about utopian systems being incompatible with AI. I would argue AI is a utopian system being used in a dystopian world where we are lacking rich usable interfaces.
We seem to dancing around a problem in the middle of the room like an elephant no one is acknowledging, and that is the interface to Artificial Intelligence and Generative AI is a place that requires several degrees of innovations.
I would argue that the first winning feat of innovation on interfacing with AI was the "CHAT BOX". And it works well enough for the 40% of use cases. And there is another 20% of uses that WE THE PEOPLE can use our imagination (prompt engineering) to manipulate the chat box to solve. On this topic, there was an article/opinion that said complex LLMs are unnecessary because 90% of people don't need it. Yeah. Because the chat box cannot do much more that would require heavier LLMs.
Complex AI and large data sets need nicer presentation and graphics, more actionable interfaces, and more refined activity concepts, as well as metadata that gives information on the reliability or usability of generated information.
Things like edit sections of an article, enhance articles, simplify articles, add relevant images, compress text to fit in a limited space, generate sql data from these reports, refine patterns found in a page with supplied examples, remove objects, add objects, etc.
Some innovation has to happen in MS Office interfaces. Some innovations have to happen in photoshop-like interfaces.
The author is complaining about utopian systems being incompatible with AI. I would argue AI is a utopian system being used in a dystopian world where we are lacking rich usable interfaces.