Human intent can mean lots of things. It really depends on the human you’re using as reference. For example if it’s a serial killer or a CEO or something…
Humans can’t align with each other. That fact makes any suggestion from them that they can align a literal alien intelligence completely ludicrous. And they aren’t stupid, so they’re just hoping most people will buy it and not pay attention. We have three directions at this point - a) AGI developed and is not aligned, things go bad in some way, b) AGI not developed but current level AI is not aligned and goes rogue in a critical way (we’ve already had some stuff happen, just not deadly yet), or c) AI stays at the current level and we manage to contain or put out the fires as misaligned actions occur, BUT human directed AI does something deadly.
There’s d) we find safe ways to do either low leve AI or AGI. I can’t see that happening, we’re aren’t being careful enough, slow enough, nor is our motive to make the safest product possible. The motive is $$$, and that only ends in a/b/c.
That seems to be SOP for any large company these days. AI is a larger scale of money dumping, but it’s the same idea. Got to get that stock number up for the investor, even if it means screwing the customers and damaging a brand name and product reputation. Line go UP!
Based on history, until it breaks, and then everyone tries to save themselves as it comes down. That’s just a financial crash though. Not sure about if AGI or whatever actually is part of the failure. Maybe watch for some insiders who suddenly duck and cover.
We’ve trained LLMs to simulate collusion, clandestine behaviour, threats, sabotage, hacking, etc… why the fuck was that the first priority lol. It’s like giving a toddler a loaded bazooka instead of a bubble blower
Hmmm… He says they will work to ensure alignment with human intent.
Humans are terrifying too.
Human intent can mean lots of things. It really depends on the human you’re using as reference. For example if it’s a serial killer or a CEO or something…
Especially billionaires like Huang.
Humans can’t align with each other. That fact makes any suggestion from them that they can align a literal alien intelligence completely ludicrous. And they aren’t stupid, so they’re just hoping most people will buy it and not pay attention. We have three directions at this point - a) AGI developed and is not aligned, things go bad in some way, b) AGI not developed but current level AI is not aligned and goes rogue in a critical way (we’ve already had some stuff happen, just not deadly yet), or c) AI stays at the current level and we manage to contain or put out the fires as misaligned actions occur, BUT human directed AI does something deadly.
There’s d) we find safe ways to do either low leve AI or AGI. I can’t see that happening, we’re aren’t being careful enough, slow enough, nor is our motive to make the safest product possible. The motive is $$$, and that only ends in a/b/c.
You’re also forgetting dumping a shit ton of money into an unprofitable pit that never meets expectations.
That seems to be SOP for any large company these days. AI is a larger scale of money dumping, but it’s the same idea. Got to get that stock number up for the investor, even if it means screwing the customers and damaging a brand name and product reputation. Line go UP!
For sure, but for how long?
Based on history, until it breaks, and then everyone tries to save themselves as it comes down. That’s just a financial crash though. Not sure about if AGI or whatever actually is part of the failure. Maybe watch for some insiders who suddenly duck and cover.
We’ve trained LLMs to simulate collusion, clandestine behaviour, threats, sabotage, hacking, etc… why the fuck was that the first priority lol. It’s like giving a toddler a loaded bazooka instead of a bubble blower