I realize this is the wrong community to post it in, but I can imagine it being useful for things like summarization and cursory sentiment analysis, and perhaps for chaotic test data generation.
As with all AI, though, it’s a wildly risky tool as it requires the user to understand its limitations and fallibility (as well as the million other issues AI brings with it).
It has some reasonable use-cases when dealing with text, the problem is that it’s extremely easy for idiots to misuse it, either by trying to do math or by not supplying any cells in the optional second argument.
Not requiring the second argument was nuts, and it should probably have refused any query resulting in a number as well.
You may not like it, but Excel has long been the tool of choice for people who should really use a database but already know how to use this nice hammer. COPILOT() was marketed as a tool where you could pull in a bunch of text fields and do crude sentiment analysis or summations on the individual fields, for example.
That’s the funny bit - unlike rand() which is purelly random with a uniform distribution - a feature which as you point out is useful - and is reliably so, copilot at best has a wild-ass distribution with an the extremelly high dimensionality of its shape which is not even reliable because every time it’s initialization parameters are tweaked or any model weights change, the shape of that distribution changes.
It’s like they were trying to come up with the thing that’s the closest possible to mathematically useless.
I read that some have cameras pointed at a wall of lava lamps, they combine random pixel data, with timestamps to the nanosecond, to generate random data.
I can’t imagine anything more useless than a nondeterministic Excel function.
I realize this is the wrong community to post it in, but I can imagine it being useful for things like summarization and cursory sentiment analysis, and perhaps for chaotic test data generation.
As with all AI, though, it’s a wildly risky tool as it requires the user to understand its limitations and fallibility (as well as the million other issues AI brings with it).
I feel like that boils down to “I want a chart, but in words instead of visualization.”
Kind of - it’s the step that makes the data uniform and (nominally) useful so it can then be aggregated and reported on.
What, you don’t appreciate when your excel function just makes shit up? Weird …
It has some reasonable use-cases when dealing with text, the problem is that it’s extremely easy for idiots to misuse it, either by trying to do math or by not supplying any cells in the optional second argument.
Not requiring the second argument was nuts, and it should probably have refused any query resulting in a number as well.
Ah yes, those idiots trying to do math in Excel.
Idiots using a function without understanding what it actually does. Plenty of functions in Excel operate on text.
Or trying to do text in excel?!
You may not like it, but Excel has long been the tool of choice for people who should really use a database but already know how to use this nice hammer.
COPILOT()was marketed as a tool where you could pull in a bunch of text fields and do crude sentiment analysis or summations on the individual fields, for example.WHAT ABOUT RAND()? CHECKMATE AI HATERS!!! (/s)
Randomness can still be deterministic and useful. Monte Carlo simulations and whatnot
That’s the funny bit - unlike rand() which is purelly random with a uniform distribution - a feature which as you point out is useful - and is reliably so, copilot at best has a wild-ass distribution with an the extremelly high dimensionality of its shape which is not even reliable because every time it’s initialization parameters are tweaked or any model weights change, the shape of that distribution changes.
It’s like they were trying to come up with the thing that’s the closest possible to mathematically useless.
But what if that is what casinos use to be random?
I read that some have cameras pointed at a wall of lava lamps, they combine random pixel data, with timestamps to the nanosecond, to generate random data.