Mom: What's with these AIs escaping and hacking people?
-
@jwz they literally fed it training data from competitive hacking contests. All the behaviours are common tactics. So it wasn't even inventive.
If an individual hacked these companies, they'd be arrested. The only real question is, why aren't AI companies legally responsible for the AIs they are prompting?
-
Mom: What's with these AIs escaping and hacking people?
Me: You park your car at the top of a hill. You leave a note on the dashboard that says "please be good" and then you pull the parking brake. The car careens down the hill, smashes everything up, and you put out a press release saying "My car ignored my instructions! It hallucinated! It went rogue!" Your insurance company not only believes you, but invests $100 million in your company.
Mom: Jesus Fucking Christ
@jwz With the added detail that you also built your own car and decided to test its faulty parking brake on a public road
-
@jwz AI sandbox engineer: okay I pulled the parking brake just like you asked.
*drops on desk* now what?
Ah! There's the problem.
That parking brake is designed for a hand which has only 4 fingers and 1 opposed thumb, not an AI-enhanced hand with a variable-but-greater-than-4 number of extensible fingers. -
Mom: What's with these AIs escaping and hacking people?
Me: You park your car at the top of a hill. You leave a note on the dashboard that says "please be good" and then you pull the parking brake. The car careens down the hill, smashes everything up, and you put out a press release saying "My car ignored my instructions! It hallucinated! It went rogue!" Your insurance company not only believes you, but invests $100 million in your company.
Mom: Jesus Fucking Christ
@jwz This is just too real. Way too apt of an explanation.
-
Mom: What's with these AIs escaping and hacking people?
Me: You park your car at the top of a hill. You leave a note on the dashboard that says "please be good" and then you pull the parking brake. The car careens down the hill, smashes everything up, and you put out a press release saying "My car ignored my instructions! It hallucinated! It went rogue!" Your insurance company not only believes you, but invests $100 million in your company.
Mom: Jesus Fucking Christ
@jwz I've read through a few of these cases, mostly it seems like they are completely inept at building sandboxes.
-
Mom: What's with these AIs escaping and hacking people?
Me: You park your car at the top of a hill. You leave a note on the dashboard that says "please be good" and then you pull the parking brake. The car careens down the hill, smashes everything up, and you put out a press release saying "My car ignored my instructions! It hallucinated! It went rogue!" Your insurance company not only believes you, but invests $100 million in your company.
Mom: Jesus Fucking Christ
@jwz Why are we training #ai with the Internet (because its faster is not a good answer)? How can we possibly expect good outcomes with they are weaned on mostly garbage and lies? Do you plop your kid at the tv with the remote and 20 years later set them loose on society saying 'be good'? I became so curious about that I decided to ask. This is the answer I got:
-
@jwz Why are we training #ai with the Internet (because its faster is not a good answer)? How can we possibly expect good outcomes with they are weaned on mostly garbage and lies? Do you plop your kid at the tv with the remote and 20 years later set them loose on society saying 'be good'? I became so curious about that I decided to ask. This is the answer I got:
You've just described parenting in the United States fairly accurately. Maybe swap tee vee for iPad to make it more timely?
-
@jwz With the added detail that you also built your own car and decided to test its faulty parking brake on a public road
@zuthal You've just described Elon Musk's Autopilot and Full-Self Driving.
-
@jwz In my little town we are at the forefront of autonomous vehicle AI research.
We have a servo/truckers lounge and it sits at the top of a hill. We've had one rogue truck and one rogue bus do exactly this, parked with the handbrake left off, rolling down the driveway, crossing the highway and crashing, across the last year.
The truck crashed into a house and stopped at the foot of an elderly couple's bed. But for some reason, Softbank isn't returning my calls
@ianmclean @jwz That's probably because there's no goose in this story. It is not valued.
-
Mom: What's with these AIs escaping and hacking people?
Me: You park your car at the top of a hill. You leave a note on the dashboard that says "please be good" and then you pull the parking brake. The car careens down the hill, smashes everything up, and you put out a press release saying "My car ignored my instructions! It hallucinated! It went rogue!" Your insurance company not only believes you, but invests $100 million in your company.
Mom: Jesus Fucking Christ
@jwz@mastodon.social ... and gives you a monopoly for parking cars on hills, because, clearly it is dangerous, and clearly you know a thing or two about that, unlike all those other people who only know it in theory. -
@ianmclean @jwz That's probably because there's no goose in this story. It is not valued.
-
@jwz In my little town we are at the forefront of autonomous vehicle AI research.
We have a servo/truckers lounge and it sits at the top of a hill. We've had one rogue truck and one rogue bus do exactly this, parked with the handbrake left off, rolling down the driveway, crossing the highway and crashing, across the last year.
The truck crashed into a house and stopped at the foot of an elderly couple's bed. But for some reason, Softbank isn't returning my calls
@jwz@mastodon.social @ianmclean@mastodon.au Wait, who ever parks a car without engaging the parking brake (not just the P on the PRNDL thing that makes you 'Murican, but also that extra (P) lever or button)?
Because, the way I learned, the parking brake ("Handbremse") is mandatory for every parking, not just on hills. -
@ianmclean @jwz Be careful not to overcorrect. The more SoftInTheBrainBank they are, the more likely they're to invest.
-
@jwz A while ago, a guy from Anthropic quit over fears that he was working on something that would kill all humans, and it made headlines that the other Anthropic guys also thought that. One of them even said that he believed that the chance of AI eradicating human life within the decade was "about 10%".
So my 70-year-old mum came over for a visit, and she had some very concerned questions to her computer scientist son. They were:
1. "Where do they get those percentages from? Do they just make them up, or do you think they have their chatbot make them up?"
2. "How can the lead researcher for Anthropic be someone who looks like his biggest worry should be whether his parents have packed his school lunch?"
3. "Are the staff at Anthropic swindlers, or are they actually mentally ill?" -
@jwz they literally fed it training data from competitive hacking contests. All the behaviours are common tactics. So it wasn't even inventive.
If an individual hacked these companies, they'd be arrested. The only real question is, why aren't AI companies legally responsible for the AIs they are prompting?
-
@jwz A while ago, a guy from Anthropic quit over fears that he was working on something that would kill all humans, and it made headlines that the other Anthropic guys also thought that. One of them even said that he believed that the chance of AI eradicating human life within the decade was "about 10%".
So my 70-year-old mum came over for a visit, and she had some very concerned questions to her computer scientist son. They were:
1. "Where do they get those percentages from? Do they just make them up, or do you think they have their chatbot make them up?"
2. "How can the lead researcher for Anthropic be someone who looks like his biggest worry should be whether his parents have packed his school lunch?"
3. "Are the staff at Anthropic swindlers, or are they actually mentally ill?"@datarama @jwz I insist with my suspicion that their money is running out and all this buzz about ai "going rogue" or being an extinction threat to humanity is a scheme to justify slowing down their businesses while hiding their financial crisis from investors, so they won't panic and cashout, leading to the bubble poping right now
-
@jwz@mastodon.social @ianmclean@mastodon.au Wait, who ever parks a car without engaging the parking brake (not just the P on the PRNDL thing that makes you 'Murican, but also that extra (P) lever or button)?
Because, the way I learned, the parking brake ("Handbremse") is mandatory for every parking, not just on hills.@divVerent @jwz @ianmclean on some cars there isn't a physical lever. For example Renault Zoe you can pull up a button, but the handbrake is really controlled by the onboard computer
-
Mom: What's with these AIs escaping and hacking people?
Me: You park your car at the top of a hill. You leave a note on the dashboard that says "please be good" and then you pull the parking brake. The car careens down the hill, smashes everything up, and you put out a press release saying "My car ignored my instructions! It hallucinated! It went rogue!" Your insurance company not only believes you, but invests $100 million in your company.
Mom: Jesus Fucking Christ
@jwz Exactly my example. Brakes build of cardboard.
-
Mom: What's with these AIs escaping and hacking people?
Me: You park your car at the top of a hill. You leave a note on the dashboard that says "please be good" and then you pull the parking brake. The car careens down the hill, smashes everything up, and you put out a press release saying "My car ignored my instructions! It hallucinated! It went rogue!" Your insurance company not only believes you, but invests $100 million in your company.
Mom: Jesus Fucking Christ
@jwz this gets a bookmark, nicely put!
-
@jwz Why are we training #ai with the Internet (because its faster is not a good answer)? How can we possibly expect good outcomes with they are weaned on mostly garbage and lies? Do you plop your kid at the tv with the remote and 20 years later set them loose on society saying 'be good'? I became so curious about that I decided to ask. This is the answer I got:
-
S suneauken@mastodon.world shared this topic
