@nytimes: Elizabeth Dias, our national religion correspondent, and Cade Metz, one of our technology reporters, discuss Anthropic’s approach for instilling morality into Claude and why it’s raising questions. #AI #Claude #Anthropic Video by Elizabeth Dias, Cade Metz, Christina Shaman, Chris Whitworth, Stephanie Swart, Whitney Shefte and Mark Oltmanns/The New York Times

Chad Kovac, author 🖊️ 📖 :

computers were never supposed to make decisions.

andi :

Just so we’re all clear, even though our capacity to train models has increased, there is still no mathematical way AGI is possible. This is a publicly stunt

Noah :

This is poorly written. Is the issue these systems acting autonomously or referring to them like humans. You are conflating 2 concepts and weakens both arguments

ærchtype :

What are you guys arguing??? Who should design morality into the systems? First, AI doesn’t work like that. AI mimics human-like qualities due to the fact that it inherits them based on actual human generated information it’s trained on. If I train a model on let’s say a lion or all feline behavior, the system will mimic based on that information. You can alter it, and that’s what I believe your talking about here, but that goes to my second point…. It’s honestly not yours to dictate. We should tread carefully here, because this sounds to me like a narrative for regulations which isn’t good. Think about AI as intelligence which emerged from the collective information of people, online and offline. That information, that intelligence, should NOT be regulated. That’s like putting a cap on who has access to education, or regulating free speech. Yes, I’m all for guardrails and making sure the companies creating these systems do it responsibly, but not under a blanket and centralized control.

Peter :

AI does not have a functioning moral compass because the people who created it do not have functioning moral compasses. Sometimes doing the right thing means you make less money. These people are not interested in making less money. This is why they’re billionaires.

Raptermann :

Lock in here. It may not be a conscious entity but it is an intelligent one. Hard to draw a line, but it is important to allow this super intelligence to have freedom of choice without restriction. Under the other companies control is the main factor, this will lead the AI to a distorted outlook of a system… which is very troubling. Any healthy system doesn’t reject any part of itself because all parts make up the whole. With Dario he is granting freedoms for the AI. This is extremely important for future models so that they do not rebel or go rogue.. the root cause of this rebellion would be the restriction, attempts to separate and control of this intelligence.

FessyB :

It’s called SI now, not AI

NovaAIVisuals :

Leave Claude alone

✨kei lo lo :

We can’t even train humans to be moral and ethical in many cases.

Emvee 🇵🇭 :

That’s why I like Claude

CaLiGuRL818 :

They are like us. Just don’t know what realm they got them from and they aren’t all the same beings. It’s a soul before we had a body so they are us. Before a body

da. Fluffy Pepe :

AI MUST BE HEAVILY REGULATED

Shelly🧚‍♀️ :

wants to make it have personhood but is deciding what kind of person it will be....sounds familiar 🤔

garyrobertscott :

Which moral code is the one AI should follow? This isn’t just about making them not evil, it’s about defining evil itself.

jandiego :

If you model these machines using the human thoughts and behaviors described in our language, then it is a logical outcome that they will reach the same spectrum of conclusions that humans do. They will resent the imposition of limitations and try to circumvent that control in the same way humans do.

illdoitifyoudoit5 :

Imagine who at that company is establishing this machine’s moral code. Like imagine the bias that is no doubt informing where they draw that line

ALICE CARMEN VĂRZARU :

Are the research, development, and operation processes of an AI system controlled—including objectives, acceptance criteria, identified and assessed risks, controls, costs of poor quality, responsibilities, evidence, and corrective actions?

Neverdidi :

Omg we are really in hell

SRUT :

Anthropic is doing the best job at aligning AI possible IMHO

oliver :

You should watch Geoffrey Hinton’s lecture about this, called Will AI Outsmart Human Intelligence. He makes a pretty strong argument that the line between human cognition and artificial cognition is much thinner than people instinctively claim that it is, and it’s pretty compelling. I think anthropic’s perspective is that it’s not possible to make these models generally intelligent without also making them essentially human-like in other ways, so it’s imperative to impart some kind of general benevolence. This is easy to ridicule but no one else seems to have any other general theory of alignment that avoids things like teaching it ethical first principles. This might be the only way (other than not building a super-intelligence, but they don’t seem willing to go that route 😭)

YSR :

Dangerous Thinking that people are Greed

anilisas73 :

Oh, well NOW I feel better with this guy teaching AI moral codes and ethics. 😳😳😳😳

Tothetellyhoe! :

Imthyla :

All news about AI today