Future Tech

there have been no incidents because it's not a problem in the slightest. the entire premise of "using ai as a friend is bad" is baseless and is just a false pretense for the underlying actual opinion of "lol you have to use ai to have any friends what a loser"

I don't think it's a problem either but I'm still open to considering it might happen, I just can't imagine what that looks like. I'd prefer examples of reported incidents.

I swear by GPS. There are numerous cases of people following GPS instructions into harms way (one example). I'm sure I could find examples of death.

this is an interesting article to use here imo, all of these examples seem to be not very recent and the main image implies the use of those old awful gps devices. those weren't particularly robust devices lol. not sure if this is a very good example overall but I get the point.

but yeah idk what it would look like either. how would one even look for examples that are actually causal? run a google search for something like "incidence rate of mental illness development in otherwise healthy individuals based on extended conversational use of predictive text models?" there seems to be more examples of these text models potentially being beneficial for various aspects of mental health

i have a very good friend whos a pink octopus. he's not ai...but he knows exactly where he is @ all times. yee !

The article was old. There are more recent ones. And that's hardly a technology one might consider human, however, they follow instructions without care for themselves.

Might AI give a user the wrong information about medications or cooking instructions (I'm just thinking undercooked meat). I expect voice assistants would always use a disclaimer about consulting a doctor. What if it was a missed reminder for medications?

I had a family member that thought every time they woke up it was time to take medications. This is fine in the morning... but if you take multiple naps a day you might take too much.

What about AI software that intentionally (through dev bias) tried to influence the user to do something? Or if the learning models were skewed/biased in some way? Tin foil hat: what if it were government intervention to influence a population.

Texting and IM has really influenced how people comunicate. Sometimes more terse. Acronyms. Less grammatically correct. And often don't even say good bye (because the chat remains there for when you come back and pick it up). If voice assistants are primarily feminine voices... will this train people to communicate with feminine voices differently and treat more people like they are AI?

I think that answers should be appended with an encrypted QRcode that identifies the question/context, the AI bot version, and the sources pulled, and that the whole package should bundled into a new "AI" file format where the QRcode cannot be (easily anyway) separated from the answer. If you want to copy and paste it, great, but it will be stamped/watermarked like a stock photo image. Stops the cheating on school projects, stops the people who use it to pretend they are creating the content, forces people to paraphrase at the very least.

This could be reevaluated later as the tech (A) gets better, but more importantly (B) when people finally learn not to trust that things they are reading are somehow more true than if Uncle Joe came up to them at Thanksgiving and told them a conspiracy theory while three sheets to the wind.

Clearly that will take quite a while, since people still haven't figured this out for the Internet at large and it's been decades now.

disclaimers about incorrect info is honestly pretty easy to implement and based on how these models work, those disclaimers can and will show up necessarily show up in the output regardless. consider the cooking example of undercooked meat. if the training data has a bunch of recipes with ingredient lists, any text about meat needing to be a cooked to specific temperature is also in the training data. devs could weight this kind of thing specifically they wanted to.

speaking of which, ai devs can (and do) bias their model for a number of different purposes. some of them aren't that objectionable imo, chatgpt for example is hardcoded to censor various racial slurs. but even then you can usually prompt engineer it around it. but some of them, however. this tweet is very explicitly political and I don't mind if you have to delete this link

you should take a gander at the tweet though and prompt engineering as a new field in infosec. I'm on a third read of this book about chatgpt, check this out too if you know your maths. imo the only option for llms to stay neutral and constantly checked by individuals who aren't privately owned companies and/or the government is requiring, at minimum, training sets and what material they contain be released with an open source license.

and to the bit about feminine voices, yep a bunch of people have noticed that default voices for siri, alexa and the like, are almost always female. some people think that reinforces stereotypes but getting into this on the forums is probably a bad idea for everyone involved lol

I wasn't trying to lead it to a gender discussion... rather, do you want to be talked to like someone talks to their voice assistant? That could be one negative social impact.

Suicide booths

Or kiosks, whichever you prefer. Kiosk sounds more modern

I know, just a preemptive since I don't wanna go there either

the way they talk could definitely be a problem, I think. fortunately tho voice assistants like siri are not even close to how verbose and convincing ai voice models can be, a lot of that probably has to do with computing resources. I splurged on a 4090 super earlier this year and can do some ai stuff locally, turns out amazon wants to keep the alexa prices low. like $50 low lol.

talking directly with chatgpt tho, not sure. if a new coworker talked with me like I was an llm trained on terabytes of human text, I'd probably just think "well he's certainly on the spectrum"

o ye of little faith

I stand by this post

✊

Is he a character in the drama "The Boys"?