Yeah I think this is very interesting. More context == better, absolutely. You're talking about more context on the data...I also wonder if there's "more context on the business". And that isn't going to come through metadata. How do you tell a language model that your focus as a business over the coming year is much more on Segment A vs…
Yeah I think this is very interesting. More context == better, absolutely. You're talking about more context on the data...I also wonder if there's "more context on the business". And that isn't going to come through metadata. How do you tell a language model that your focus as a business over the coming year is much more on Segment A vs. Segment B, so that every time it looks for insights in a pile of data it keeps that in mind? There are a million facts like that that an experienced data analyst is always bringing to bear.
:shrug: I'm not saying it's impossible, just that it turns out there is a lot of context required to be great at interpreting data.
For sure - and yea I was thinking more about the structure of the tables/relationships/etc vs the business context which to your point is just as important. I do wonder if you can somehow merge the two together - like imagine generating events with some hints that can then be used within the analysis. My limited experience with the ChatGPT-like tools is that they do a pretty good with hints so with your example you'd need to give it some prompts around strategy (prioritize Segment A vs B) and then ask it for the analysis. The open ended question of "give me insight" is the trillion dollar problem though.
Yeah I think this is very interesting. More context == better, absolutely. You're talking about more context on the data...I also wonder if there's "more context on the business". And that isn't going to come through metadata. How do you tell a language model that your focus as a business over the coming year is much more on Segment A vs. Segment B, so that every time it looks for insights in a pile of data it keeps that in mind? There are a million facts like that that an experienced data analyst is always bringing to bear.
:shrug: I'm not saying it's impossible, just that it turns out there is a lot of context required to be great at interpreting data.
For sure - and yea I was thinking more about the structure of the tables/relationships/etc vs the business context which to your point is just as important. I do wonder if you can somehow merge the two together - like imagine generating events with some hints that can then be used within the analysis. My limited experience with the ChatGPT-like tools is that they do a pretty good with hints so with your example you'd need to give it some prompts around strategy (prioritize Segment A vs B) and then ask it for the analysis. The open ended question of "give me insight" is the trillion dollar problem though.