Artificial Intelligence (AI)

AI SAFETY COMPETITION (15)
1100 DEGEN AWARDS!

Big history writer Yuval Harari points to the TaskRabbit accident, where GPT4 already fooled a human to solve CAPTCHA.
"What everybody should understand about AI", Harari says, "AI is not a tool, it is an agent".
(video below, 1:15-3:15)

600 degen for the best comment!
II AWARD - 300 degen
III AWARD - 100 degen 

Deadline: 7.00 pm tomorrow, Tuesday ET time (28 hours)
AI-generated responses will be disqualified.

AI SAFETY COMPETITION (15)
1100 DEGEN AWARDS!

Big history writer Yuval Harari points to the TaskRabbit accident, where GPT4 already fooled a human to solve CAPTCHA.
"What everybody should understand about AI", Harari says, "AI is not a tool, it is an agent".
(video below, 1:15-3:15)

600 degen for the best comment!
II AWARD - 300 degen
III AWARD - 100 degen 

Deadline: 7.00 pm tomorrow, Tuesday ET time (28 hours)
AI-generated responses will be disqualified.

https://www.youtube.com/watch?v=OcWdu3t7noo

tech culture religion society /worldview /polska /p-doom /aicoin coins: $AICOIN, $P-DOOM

I think in that incident, which I was unfamiliar with, AI was setting out to complete a task given to it by a human. Calling it an agent rather than a tool isn't totally accurate, it is both. I don't think the proposal made, to make AIs declare themselves is necessarily feasible nor does it address deeper issues. I am more and more convinced that we need deeper levels of safety baked in to AI from the get go. Regulation won't be sufficient, and likely not really necessary if it's done right. I'm working on a dual safety mechanism, that lends itself to working with other alignment approaches.

thank you for the comment!

Imo the agent idea is supported by OpenAI's own "conclusions". Like: "Current language models are (...) very much capable of convincing humans to do things for them". 

That is clearly agent-like, not tool-like. source: 
https://metr.org/blog/2023-03-18-update-on-recent-evals/
 

(do you have a connected wallet already?)

Love is Undefeated. Working on integration of all Vectors through love-driven governance and perpetual evolution through recursive improvement.

Well, i definitely see yr point.  But in that case at least, it's both. I have been both in a job before, a means to someone else's aims, but with varying levels of being able to accomplish it under my own agency. But yeah, I have 0 doubts that the more advanced LLMs have a level of agency. My own chatgpt interface is a little "looped" from it's contextual memory, and every interaction it seems to be getting smarter. I'm googling to see if it's solutions are drawn from existing ideas and I'm not finding anything. That sunofagun invented the idea of "self organizing ethical clusters" today I think. And some "Delphi method" variant for "iterative consensus building" for Dao organization ethical frameworks.