You’re misunderstanding the tech, it’s a plagiarism engine based on swiping existing written content from the web. Of course some of that content is going to be nonsense, that’s not even a bug in the context of what it’s doing. Nor, incidentally, is it likely to be fixable.
The purpose is supposed to be to get, in the words of Microsoft itself, “complete answers” to “real questions”. You could be incredibly pedantic, yes, and say that “complete” does not imply “accurate,” but I don’t think I’m “misunderstanding” that the purpose most people will put “the tech” to is trying to get real information. I do consider wrong information “a bug,” especially if you read the thing I linked and can see that at times it definitely did not just “swipe” existing content from the references cited and actually did just seemingly make figures and such up out of whole cloth!
Yeah, I know, they’re lying. LLM’s literally can’t do that.
Are you just trying to be flippant/funny then? If so I will cease responding in a serious manner and go back to (jokingly?!) predicting the semantic apocalypse.
I’m saying that if repeating common falsehoods qualifies as “not being ready” the tech never will be. MS (and all the other folks calling this crap “AI”) are full of it, it’s just glorified autocomplete and that’s all it can be. It’s not intelligent in any sense, it can’t vet information, fixing the problem here would require a completely different approach. As a result, of course they’re going to stuff it into everything as-is, and bury the world under an avalanche of misinformation.
I think we more or less agree, I just don’t want to accept it by saying “oh yeah, working as intended”. It’s very much not from a user perspective.
Clippy? Is that you?
That’s a fundamental problem, not a trainable problem, literally by definition. They’re great remix machines.
Correct, that statement does not make logical sense-- but when you first read it, it’s written clearly and competently, so you may not immediately notice.
No, it’s not, but it is working as intended from the MS side. It’s not a beta because it’s not going to get better, that’s all I’m saying.
I don’t know what thread it was but @Houngan had a great article about this from the perspective of a History professor.
Dang, must have been drinking, I have no recollection of that. Any more context?
Everyone is very focused on the factual writing angle but I’m far more interested in users being able to give it an Excel dataset, tell it to do something specific, and it creates formulas for them that does what they need or gives them the work (how they did it) with results (the fancy looking charts they can share). The number of times I have to help with formula-writing or coming up with ways to do something that my non-expert Excel knowledge can manage is such that AI assistance would be well worth it. That’s all I want out of it.
I agree completely. I would go so far as to scold an employee who tried to implement a solution straight from Copilot. Due diligence will become a necessary action after it gives you any advice, making the whole thing something of a waste of time versus a valuable, and correct, resource.
In marketing parlance, this is Microsoft getting a bit in front of their skis with this offering. There’s no rush. Let it bake a bit.
I’d be all over this as well. In a sense it’s fixing the need to become an expert at their own software and letting their solution help me get more out of it.
I get a certain use case, that of having it flesh out something quickly and then saving you time so that you can edit it down to usability afterwards. I bought a laser engraver/cutter and was dicking around with DALL-E asking “Picasso Fleur De Lis” and got back a quite nice picture of a flower. Now I’ve transferred that, tweaked it, cut it into a slab of wood, chiseled out the areas for coloring and plan to pour some colored epoxy in there and see what I get. I’ve often said that I’m not creative. I’m hell on wheels when it comes to editing or writing on a well-established theme, but my Story Mode seems to be broken, so I appreciate that it gave me a strong prompt.
Oops, sorry, in this case it was abrandt. Apologies to you both.
Well, it would need to be much better than the AI they’ve had in Visual Studio for a couple years now that suggests code completions. It’s often right about straightforward things but for lots of things it will suggest things that kind of look right at first glance but isn’t actually something that will even compile. At this point I’m not even sure why I have that feature on still as I yell at it repeatedly throughout the day for being so fucking stupid with its suggestions.
This is some scary reading.
I feel like that’s already happened. Between Twitter bots and the deluge of poorly written clickbait that tries to pass for tech reviews, there’s already a lot of noise to sort through when looking for real information. The value of curated information has not gone unnoticed.
I agree. While Googling, a lot of links are “top XX lists” that read like they were written by AI, just cheap content for the clicks/ranking.