Lots of chatter about the AI that told its user to die...
Now, this is a perfect time to be skeptical. Should we believe it? It's a common game to trick these AI tools into saying stupid things, it's even got a name: prompt engineering, and if you're any good at it, you can get a job doing it.
Now, this is a perfect time to be skeptical. Should we believe it? It's a common game to trick these AI tools into saying stupid things, it's even got a name: prompt engineering, and if you're any good at it, you can get a job doing it.
Now, if we wanted to figure that out, how would we? Well, we could ask Google, the erstwhile owner of this alleged genocidal bot. Obviously Goo-do-no-evil-gle will not want the bad press and will cover up. Oh, here it comes... tomshardware.com/tech-industry/…
It's not a trivial question! Recall that Google does secret work for the Israeli defence forces, and likes to do that work, so there's a disturbing correlation here, and we hope there isn't a causality.
How to find out?zerohedge.com/technology/you…
Presumably google will also have the logs of this particular trace and be able to show the prompt engineering that led up to this outcome. But will they share it with us? Probably not. If they do share it, will we trust it?
We have a dilemma. On the face of it, we can't trust the user and we can't trust the owner. How would we solve this?
There is a way...
We have a dilemma. On the face of it, we can't trust the user and we can't trust the owner. How would we solve this?
There is a way...
strange to say, I know, but to solve this problem of AI ethics, we could use the blockchain. In short, the prompts should be pushed onto the blockchain, as is the engine itself. And of course the outcome of the prompt.
With those three datums {input, output, machine} google could prove to us what happened.
(And, extra points if it is open, as then we could prove it ourselves. OK, insert in here some blabla about zero knowledge proofs and homomorphic encryption and ... just to make this post seem credible!)
With those three datums {input, output, machine} google could prove to us what happened.
(And, extra points if it is open, as then we could prove it ourselves. OK, insert in here some blabla about zero knowledge proofs and homomorphic encryption and ... just to make this post seem credible!)
If we're serious about the ethics of AI and where we're going ahead with pasting the planet with it, then we do rather need a solution to this problem. Putting the AI onto the blockchain might not be the total answer, but it's a bloody good first step.
If you like this notion, you can read more at mdpi.com/1999-5903/11/8… which was the original paper that combined AI & blockchain.
If you like this notion, you can read more at mdpi.com/1999-5903/11/8… which was the original paper that combined AI & blockchain.
Also in Chinese! Here:
thanks to the unstoppable @lochaiching!lochaiching.medium.com/%E7%BF%BB%E8%A…
thanks to the unstoppable @lochaiching!lochaiching.medium.com/%E7%BF%BB%E8%A…
@lochaiching @threadreaderapp unroll please
• • •
Missing some Tweet in this thread? You can try to
force a refresh
