Nope, it's about human flourishing being an optimum of almost-zero utility functions and gradient descent looking like a terrible way to control inner preferences.
SAFETY
-
Detecting Additional Forces Beyond Token Prediction in Fine-tuned LLMs
By
–
Many LLMs have already been RLHFed and finetuned into activities other than "predict the next token a human would write". This being the case, how would you tell if the output was being driven by some extra force above that and all the finetuning?
-
Moral and Ethical Aspects of Artificial Intelligence
By
–
The moral and ethical aspects of Artificial Intelligence! https://
hindustantimes.com/ht-insight/fut
ure-tech/the-moral-and-ethical-aspects-of-artificial-intelligence-101695368354776.html?utm_source=twitter
… #Robotics #robot #OpenAI #bots #KillerRobot #roboticsainews #opensource #Python #tech #technology #cobot #humanoid #ML #Algorithms #Robotech #RobotDesign #Engineering #AIEthics #EthicalAI #AI -
Concerns about artificial general intelligence development and risks
By
–
This is the whole artificial general intelligence situation but hey I'm just overreacting right?
-
TESCREAL bundle influences catastrophic risk and politics discourse
By
–
Yep the #TESCREAL bundle is running the show. H/t @xriskology "Leigh will borrow the estimate of Oxford philosopher Toby Ord in a speech on Friday exploring the intersection between catastrophic risk and extreme politics” https://
theguardian.com/australia-news
/2023/sep/22/one-in-six-chance-of-a-species-ending-event-in-next-century-labor-mp-andrew-leigh-warns
…. -
Confrontation with Shane Legg over paper misrepresentation claims
By
–
Well. I now have a guess for where they got this from, and Shane Legg is out here telling me I’m misrepresenting his paper. The nerve. https://
arxiv.org/pdf/0712.3329.
pdf
… -
Eliezer Yudkowsky denies wanting to control ASI
By
–
Who told you the lie that I wanted to centralize ASI or control it? I want it to not exist. I cannot control it. It kills everyone regardless of who thinks they control it.
-
Nuclear Weapons and AI: Understanding Different Threat Categories
By
–
They are of course vastly different. As Marvin Minsky observed, nuclear weapons are not really dangerous because nuclear weapons are not self-replicating. To this I would add that artificial pathogens are not really dangerous because they're not smart.
-
AI Agents: Future of Autonomous Goal-Achieving Systems
By
–
Imagine an AI that can understand your high-level goals and use all its tools and resources to achieve them—talking to people, to other AIs, and anything else it needs. That's the future of AI, and it's closer than you think.