That works well when you have time and the right humans with the right knowledge happen to be frequenting the space.
In a lot of cases, I have seen that system used to offload the research burden onto somebody else.
Those spaces have been increasingly taken over by bots, astroturfing, or armchair ‘experts’ presenting mis/disinformation or parroting something they saw on Youtube out if context.
Granted, a lot of LLMs are also trained on this same broken information.
For more abstract concepts, though, it gets to be a bit trickier.
One of my main advantages using LLMs has been for electronics repair. Without a schematic, I’m sometimes limited to obscure markings on a part that I need to replace.
I used to sink quite a lot of time combing through datasheets trying to deduce a match, with LLMs that already have those datasheets integrated I inout the marking and form factor to get connected to a limited set of candidates to verify for myself. Hell, in some cases a picture of a half-blown component has gotten me where I need to be.
I just don’t need that level of depth for every query, when I want tacos near me a map with indexed keywords is more than enough.
That works well when you have time and the right humans with the right knowledge happen to be frequenting the space. In a lot of cases, I have seen that system used to offload the research burden onto somebody else. Those spaces have been increasingly taken over by bots, astroturfing, or armchair ‘experts’ presenting mis/disinformation or parroting something they saw on Youtube out if context.
Yeah, this is why smart, community informed architecture is so important and difficult to get right, and it is also why corporate social media is a terrible idea/business if you care about the health of the communities on it.
To be honest though, old reddit was a great source of information before monetization became the focus, so is the fediverse, yes you have to be extremely skeptical of course you do but in my experience my interactions on the fediverse the vast majority are clearly with human beings.
I just don’t need that level of depth for every query, when I want tacos near me a map with indexed keywords is more than enough.
Here you expose how little you have thought about the constant need for human curators/gardiners of spatial GIS data, there is no robotic shortcut, human beings have to maintain those systems with accounts that have histories that establish their general intent is genuine, which doesn’t mean you can trust the map 100% but it does mean it wasn’t hallucinated out of thin air by an AI.
The natural world does not come with a quantified representation of itself, but we are constantly awash in so much data that we forget that it must come from someone with their own interests, flaws, and ideologies, no matter what they claim, or how much computation they put it through.
Here you expose how little you have thought about the constant need for human curators/gardiners of spatial GIS data
Not sure what I wrote to lead you to this conclusion.
My intention was mostly to highlight and juxtapose areas where an “AI” system can be superior to a human one (combing through thousands of datasheets quickly) to another where the human curated option is ideal without “AI” involvement (maps and tacos)
The takeaway that I didn’t spell out being that I don’t think it should be automatic / compulsory for every query. Users should, at a minimum, have a dedicated extra interaction before their query is parsed by LLM.
That works well when you have time and the right humans with the right knowledge happen to be frequenting the space. In a lot of cases, I have seen that system used to offload the research burden onto somebody else. Those spaces have been increasingly taken over by bots, astroturfing, or armchair ‘experts’ presenting mis/disinformation or parroting something they saw on Youtube out if context.
Granted, a lot of LLMs are also trained on this same broken information.
For more abstract concepts, though, it gets to be a bit trickier.
One of my main advantages using LLMs has been for electronics repair. Without a schematic, I’m sometimes limited to obscure markings on a part that I need to replace. I used to sink quite a lot of time combing through datasheets trying to deduce a match, with LLMs that already have those datasheets integrated I inout the marking and form factor to get connected to a limited set of candidates to verify for myself. Hell, in some cases a picture of a half-blown component has gotten me where I need to be.
I just don’t need that level of depth for every query, when I want tacos near me a map with indexed keywords is more than enough.
Yeah, this is why smart, community informed architecture is so important and difficult to get right, and it is also why corporate social media is a terrible idea/business if you care about the health of the communities on it.
To be honest though, old reddit was a great source of information before monetization became the focus, so is the fediverse, yes you have to be extremely skeptical of course you do but in my experience my interactions on the fediverse the vast majority are clearly with human beings.
Here you expose how little you have thought about the constant need for human curators/gardiners of spatial GIS data, there is no robotic shortcut, human beings have to maintain those systems with accounts that have histories that establish their general intent is genuine, which doesn’t mean you can trust the map 100% but it does mean it wasn’t hallucinated out of thin air by an AI.
https://theluddite.org/post/spectacle-of-computation.html
Not sure what I wrote to lead you to this conclusion.
My intention was mostly to highlight and juxtapose areas where an “AI” system can be superior to a human one (combing through thousands of datasheets quickly) to another where the human curated option is ideal without “AI” involvement (maps and tacos)
The takeaway that I didn’t spell out being that I don’t think it should be automatic / compulsory for every query. Users should, at a minimum, have a dedicated extra interaction before their query is parsed by LLM.
this isn’t interesting