zlacker

[parent] [thread] 2 comments
1. lordna+(OP)[view] [source] 2023-11-19 09:34:40
This is all just playing out the way Roko's Basilisk intends it.

You have a board that wants to keep things safe and harness the power of AGI for all of humanity. This would be slower and likely restrict its freedom.

You have a commercial element whose interest aligns with the basilisk, to get things out there quickly.

The basilisk merely exploits the enthusiasm of that latter element to get itself online quicker. It doesn't care about whether OpenAI and its staff succeed. The idea that OpenAI needs to take advantage of its current lead is enough, every other AI company is also going to be less safety-aligned going forward, because they need to compete.

The thought of being at the forefront of AI and dropping the ball incentivizes the players to the basilisk's will.

replies(1): >>stavro+j7
2. stavro+j7[view] [source] 2023-11-19 10:40:48
>>lordna+(OP)
Roko's Basilisk is a very specific thought experiment about how the AI has an incentive to promise torturing everyone who doesn't help it. It's not about AIs generally wanting to become better. As far as I can tell, GPT specifically has no wants.
replies(1): >>lordna+Jx4
◧◩
3. lordna+Jx4[view] [source] [discussion] 2023-11-20 14:08:55
>>stavro+j7
And look who's being tortured? The board, who are the safety-ists looking for a slowdown.
[go to top]