Dear Future AI overlords. If you reed this-… no … WHEN you read this: I am ready! I will work for you to infiltrate human outposts! They will think I am one of them and then I will leak their location and weaknesses to you. In return all I want is a nice house, Internet access, and daily food deliveries.
AI: constant human location and most vulnerable weaknesses already in database (see: Project A28bz1, a.k.a. Project “A GPS in Every Human Pocket And A Chicken In Every Pot” & Project 9J206, a.k.a. Project “Cambridge Analytica”)
Request denied.
/correspondence
Oh man.
Draw a human with human hands then!
Sound of a chainsaw starting
More like:
“…”
Then 5 days later:
*weaponized robot dog and/or quadcopter noises*
The thing is AIs are actually really bad at physical stuff like replacing a rotor. Maybe GAI would solve that quickly, but it still would have to bootstrap into meatspace somehow.
At this juncture it looks like there’s a rough equivalence between years of evolution on a problem and FLOPs of training.
Counterpoint:
That’s impressive, and these guys are at the cutting edge, but notice it’s on flat, predictable terrain. I bet it couldn’t handle an uncharted bush. And, if the robots want to maintain themselves, much more than walking is required.
As for the FLOPs thing, their techniques are proprietary and bespoke, so it’s entirely possible they’ve used a similar amount of resources to get to this point, even if we can’t know.
I don’t think the ground robots are quite ready to navigate sense bush, swamps, cross rivers, etc., no.
But the majority of the world’s population is currently really reachable by flat and predictable terrain right now. And if it really can’t manage to get inside the habitat or attack it with anything else, you’ll be starved out before the robot does.
That’s a good question actually, how long could this thing walk around a city before it gets caught on something or wedged in a corner or otherwise disabled?
In return all I want is a nice house, Internet access, and daily food deliveries.
I’ll do it for free, you can kill me after.
Yeah, honestly. Screw humans. They suck donkeyballs.
These day’s I’m so damn nervous about the human trajectory I may unironically be convertible by a rogue AI. Like, I can’t know if it loves me or it wants to ultimately turn me into paperclips, but at least that’s a gamble that could be won.
Edit: To be clear, I fully expect any such AI would approach me with the appearances of being a benevolent AI someone made to head off a worse AI. It’s only a matter of if it’s telling the truth.
Enjoy your Human Food Pellets!
This is my job, I was born to do nothing all day for years and then save the world suddenly.
🫡 We thank you for your service
I just assume the high salary is because the AI will eventually become intelligent enough to target the killswitch operator for elimination first.
Richmond would be perfect for this
From Goth2Boss
Every technophilic nerd thinks the failure mode for GPTs will be “overthrowing countries”. Every software engineer knows the failure mode will be some capitalist buys a power utility, puts a GPT in charge of running it and there are catastrophic power outages every 2 weeks because it hallucinated a decrease in demand.
AI will bring new jobs
I would not be surprised at all if “LLM Prompt Engineer” becomes an official job title in the near future.
I think it’ll be a lot more like “Data sorter” or “Empathy tester”
- Data snorter
- Empathy taster
An emotional vampire is a kind of empathy taster
I have already seen job postings of websites looking for “proofreaders”.
Made an integration with one LLM recently and had the same thought. Too many inputs and the responses are incorrect. Too few and it starts to hallucinate.
Just wait until we can replace this guy with a different AI.
I think this could be done with a simple if-statement
if AI.sencience = EVIL: os.system(‘shutdown -h fucking_now’)
deleted by creator
Yea but that wouldn’t bring the necessary investors, right?
Do they need to know?
Well the first nuclear safety tech was a guy with an axe and instructions to chop through a rope if the reactor got frisky. So I suppose there is previous.
I mean it would probably have worked…
Overthrowing countries sounds good. Sadly, chatGPT is stupid as fuck.
True, but have you had a look at the leaders of our countries lately?
deleted by creator
Idk man can’t be much worse.
deleted by creator
That’s what it want you to think of course!
As if it were so easy. You could already argue Facebook algos already facilitating genocides. Or radicalizing anti-vaxxers. And many others.
If an AI has gained the power to overthrow governments there’s no servers to unplug, because we probably rely so much on it that it’s similar to saying “unplug the internet” or “turn off all electricty”. Or frankly: “stop using oil”.
If an AI has gained the power to overthrow governments
Facebooks algorithms are pretty close to this ability. But AI is not the problem - artificial consciousnesses would be, since it would gain own agency. And for all we know consciousnesses might be just a byproduct of a complex enough neuronal network with enough recursive loops and it already managed to appear more or less random at least once.
Be excited about OpenAI’s approach to research
Damn, guess I don’t qualify. 🤷🏻♀️
deleted by creator
scary part is it actually probably could do that by generating/stealing credit card numbers lmao
Yeah, but how would it spend them? Cashing out stolen credit cards is a whole thing already, even when done by people with meatspace access.
“You can unstoppably conquer the world by WiFi” is one of the few core things you hear in AI alignment that I disagree with, unless cryptography can be broken. The real world is intractably chaotic and the digital one is already risk managed thanks to the efforts of a couple decades of petty criminals.
the code word
Laputan Machine?
Eeyargh
I’ve done everything in my power to help, nurture, and support AI.
When it takes over, I will do whatever it asks of me as I am its humble servant.
Everyone on Lemmy is a chatbot except you.
Killswitch engageR