The followup to ChatGPT is scarily good at deception (www.vox.com)
from jeffw@lemmy.world to technology@lemmy.world on 15 Sep 2024 03:39
https://lemmy.world/post/19782806

#technology

threaded - newest

mvirts@lemmy.world on 15 Sep 2024 04:31 next collapse

Imho no amount of paywall or legislation protects us from a dangerous model. It’s software that will eventually become widely available.

Hux@lemmy.ml on 15 Sep 2024 04:36 next collapse

Sooooo…the next update to AI is sociopathy:

Because AIs don’t share common human values like fairness or justice — they’re just focused on the goal they’re given — they might go about achieving their goal in a way humans would find horrifying.

Arbiter@lemmy.world on 15 Sep 2024 04:56 next collapse

A computer has no sense of right or wrong, only pass or fail.

Hux@lemmy.ml on 15 Sep 2024 13:19 collapse

Working from the definition of sociopathy, I think you could substitute “AI” in a few spots and end up with a nearly equally-accurate definition of AI:

A sociopath is someone with antisocial personality disorder (ASPD), a mental health condition that involves a lack of regard for others’ feelings and rights. People with ASPD may: Lack empathy and remorse Manipulate others for personal gain Behave impulsively or aggressively Break rules or laws Feel little guilt for harming others Seem charming at first Have difficulty understanding others’ feelings

Grimy@lemmy.world on 15 Sep 2024 15:02 collapse

A computer lacks human emotions, more at 6

Toes@ani.social on 15 Sep 2024 05:52 collapse

Are you familiar with the paperclip problem?

The idea that if you task a sufficiently advanced AI with making paperclips it’ll inevitably turn the universe into a collection of paperclips when that is its only goal.

damirK@lemmy.world on 15 Sep 2024 07:01 collapse

www.decisionproblem.com/paperclips/

Toes@ani.social on 15 Sep 2024 07:35 next collapse

Well that’s 30mins of my life I won’t get back. Hilarious clicker game.

Blaster_M@lemmy.world on 15 Sep 2024 17:32 collapse

And a spelling mistake on an ai research paper (papercloppers instead of paperclippers) spawns an entire mlp fanfiction where a game dev accidentally makes an AGI to run an MMO that ends up turning the world into a singularity. Friendship is Optimal, after all.

wendrul@lemmy.world on 16 Sep 2024 20:45 collapse

<img alt="" src="https://lemmy.world/pictrs/image/1df88d0e-6f75-444b-9b82-16dbabe965a2.png">

Give me more paperclips or at least my time back

RagingSnarkasm@lemmy.world on 15 Sep 2024 04:37 next collapse

No, I’m not.

DarkCloud@lemmy.world on 15 Sep 2024 04:56 next collapse

Good, it’s the only reliable sign of intelligent self-awareness there is, to the point that all children progress through it, starting out as bad liars, and getting better at it.

LLMs however might just be stupid, or stocasticaly incorrect.

shalafi@lemmy.world on 15 Sep 2024 13:30 collapse

I was honestly concerned my daughter might be a sociopath. Then I realized she was just 3.

db0@lemmy.dbzer0.com on 15 Sep 2024 09:14 next collapse

The peeps focusing on finding out scheming prompted an llm to generate scheming. Yawn This is the only surprising if you don’t know that llms are fancy autocompletes.

DrCake@lemmy.world on 15 Sep 2024 11:28 next collapse

This has nothing to do with the article it’s self, but I absolutly hate GIFs as the thumbnail. I get that they do it for the attention, but if every site starts doing this, my feed will look awful.

conciselyverbose@sh.itjust.works on 15 Sep 2024 13:09 next collapse

That’s ugly as hell too.

higgsboson@dubvee.org on 15 Sep 2024 14:04 collapse

Yeah, I’d disable that if I could. I’d much rather have no thumbnail than an animated one.

technocrit@lemmy.dbzer0.com on 15 Sep 2024 15:52 next collapse

Computer programs don’t deceive. They respond programmatically based on input given.

Any perceived deception by the computer is actually just irrational expectations, etc. by the user.

it will start by telling you it’s “thinking.” After a few seconds, it’ll specify that it’s “defining variables.” Wait a few more seconds, and it says it’s at the stage of “figuring out equations.” You eventually get your answer, and you have some sense of what the AI has been up to.

The opposite is true here. We are being intentionally misdirected by misleading/humanizing language away from what the computer is actually doing. Not even close to understanding what the “AI” has been computing.

However, it’s a pretty hazy sense. The details of what the AI is doing remain under the hood. That’s because the OpenAI researchers decided to hide the details from users… In other words, we’re not sure if Strawberry is actually “figuring out equations” when it says it’s “figuring out equations.” Similarly, it could tell us it’s consulting biology textbooks when it’s in fact consulting comic books. Whether because of a technical mistake or because the AI is attempting to deceive us in order to achieve its long-term goal, the sense that we can see into the AI might be an illusion.

The author conflates the actual deception of the developers with the imaginary deception of the “AI”. This type of terrible coverage is completely normal inside the “AI” bubble.

KairuByte@lemmy.dbzer0.com on 15 Sep 2024 17:55 collapse

It’s such a dumb take too. Let’s look at other things that need time to load: video games. Often times the loading screens tell you nothing about what’s actually going on, but hilariously some say things like “weeding the garden” or “sniffing some glue” or other things to that effect.

“It’s lying to us” is such a… naive take. It’s just spitting out preprogrammed text to essentially say “nothing had gone wrong but we are still working on this.”

rsuri@lemmy.world on 15 Sep 2024 16:04 next collapse

Whenever I see these articles, I feel like it’s part of OpenAI’s marketing strategy to use the fear-based media to pump up the sense of how revolutionary their tech is. Kinda like how they marketed GPT-4o by openly tying it in to the movie Her, as if ChatGPT is gonna replace relationships. They’re clearly very aware of what gets attention and going for that.

UniversalMonk@lemmy.world on 15 Sep 2024 16:39 collapse

Whenever I see these articles, I feel like it’s part of OpenAI’s marketing strategy to use the fear-based media to pump up the sense of how revolutionary their tech is

Yep, I think that’s what’s going on too.

noxy@yiffit.net on 15 Sep 2024 18:01 collapse

Just like its salespeople