@Sweetshark@Dave3307@rodhilton Or more likely, it'll be like translation software when you translate from one language to another and then back. It'll kinda get most of the gist, but some key important stuff will be garbled in the middle.
@bornach@JessTheUnstill@Sweetshark@Dave3307@rodhilton Oh, that's not my concern - my concern is that they start getting things wrong (Or different than originally posed), so that stuff appears to be hallucinating details that aren't in the original post when summarizing up things.
@rodhilton the interesting question is: are the summaries at least semantically equivalent? i can easily imagine scenarios where some ambiguity leads to differences between them, and then we will ask whose responsiblity it is. in other words: who has to check the factual correctness of the AI output?
@rodhilton
This is indeed bizarre.
Myself, I would like to work with directness and brevity, but many if not most of the emails I write go to people whom I don't know and who do not have the same familiarity with the subject matter as I do (think tech support, but different). I have to be very clear and unambiguous, and I need to explain things and circumstances as I go. This invariably leads to some length. I'd rather have it brief and direct, but that would cause more questions than I'd like.
@rodhilton I don't think that is as mad as you are implying. People use language for more than just transmitting a single message: it embeds social signals for things like friendliness, status, care, professionalism. A one line email may be read as curt, dismissive, even aggressive, stoking flame wars.
Knowing how to craft messages to include those social signals is difficult, particularly for neurodivergent people, and they take longer to write. So a tool to help makes sense.
@rodhilton I was pretty sure I saw a webcomic that basically predicted this. In the first frame, somebody wanted to sound smarter and like they'd put more work into an email, so they asked an AI to expand it into something more-verbose. In the second frame, the recipient didn't want to read such a long email, so they asked the AI to summarise it for them.
When I first saw it, I felt inspired and considered writing a blog post about how, through being a compatibiity layer, the interface between different AI agents in the future might well be... natural human language. Which is crazy. Then I remembered I actually already WROTE that blog post back in 2020: https://danq.me/sandwichware
Anyway: I wanted to share said comic with you, in case you hadn't seen it, but this morning I couldn't find it again. My first thought was that I was written by @ZachWeinersmith, so I started trying to find it. I tried using my favourite Web search engine, scoped to SMBC, but I wasn't finding it, so instead I started poking around the edges of the site to try to find a search interface. It has tags, so there's something LIKE one. Then I found a minor XSS vulnerability.
Anyway, the short of it is that now I'm neglecting my breakfast (and my dog, who wants a walk) on a Sunday morning... because I'm writing a proof-of-concept for a security issue that I found while failing to find a comic that almost inspired a blog post that it turned out I'd already written but thought of when I saw your toot.
@rodhilton this is why I view all this as farce.
No, we do not need more text. We need shorter and more precise information, promptly, not paragraphs of useless word salad. IT systems NEVER had issue with expressing themselves. NEVER. What needed improvement was UX and very precise signaling.
Word salad machine is NOT good UX and definitely NOT good signaling.
This is bullshit at all levels.
@rodhilton There's a bit in the middle of David Foster Wallace's "Infinite Jest" where he describes an entire cottage industry of photorealistic face masks designed so that people on video calls with each other could hide behind them and not ever have anyone see bedhead or skin blemishes and as time goes on it's seemed more and more prescient.
@rodhilton as is usual in tech these days, pointless Innovation and features nobody asked for and nobody really needs. Just because they can’t think of anything better or a practical use for whatever they’re doing…
Many companies have that problem in recent years.
Add comment