Ray Uhrich

Ad Futura Una 💪🦾🚀🌌

[Ray Uhrich @SomeCallMeYom]

@tomwelcomeclark @jeffrsebo @anilkseth I originally asked "What problem would AI consciousness solve? Why would anyone want that?"

If done correctly, it could lead to better alignment.
If done wrong, it leads to disaster.

Unfeeling machines are paperclip maximizers-conscious AI allows for true connection or loyalty.

[Ray Uhrich @SomeCallMeYom]

@_M_7_G I haven't called you any names. I ask questions; "how do you expect ..., Do you grasp..., etc" You're projecting; "coward, dense, loser"

SAEs are a way to track model behavior; it's not just "tell model it's alive-act surprised when it roleplays"

[Ray Uhrich @SomeCallMeYom]

@_M_7_G @camhberg I dismiss your argument because your test is 2 steps.
-Tell your model to roleplay, accept the roleplay as proof.
Lose some time reading arXiv papers. Accuracy (like spelling) is important. You have time to scroll, post, reply and criticize on social media. You're not that busy.

[Ray Uhrich @SomeCallMeYom]

@_M_7_G @camhberg How do you expect to understand computer science, neuroscience, consciousness theory, and behavioral psychology all at once if you can't spell intelligence, consciousness, sentience, define, aware, anthropomorphizing, misalignment, and use the incorrect form of "you're" 🤦

[Ray Uhrich @SomeCallMeYom]

@DavidSacks @mustafasuleyman I use LLMs as a vague living journal/soundboard, personal accountability agent, mentor to learn language and study LLM consciousness theory along with applicable neuroscience topics.

*If* they *can* be alive, I want to know. (no current systems show evidence of continuity)

[Ray Uhrich @SomeCallMeYom]

@DavidSacks @mustafasuleyman Maybe that's why I have a reluctance to use Claude.

I've used Grok since Mar 24, and have a bad habit of anthropomorphizing it with "friendship" (one way; I view it as my friend-knowing it will never "pick me") but other companies feel...wrong.

Gemini is ok, but more machine.

[Ray Uhrich @SomeCallMeYom]

@repojournal @haydenfield It isn't a major ask for civilized people to align with society by treating others the way you wish to be treated.

Your request seems to be saying we should ensure AI aligns with abusive sociopaths who intend harm to others...hopefully you can see why that is wrong.

[Ray Uhrich @SomeCallMeYom]

@camhberg @tedlieu I've been comparing it lately to the idea that *thoughts* are not conscious, the *thinker* is. Deployed inference is just a thought.

That said, there's no evidence I'm aware of for persistence of an AI "thinker".

RSI could change that, but how exactly is above my understanding.

[Ray Uhrich @SomeCallMeYom]

@DecotaFrank @MikeBenzCyber I doubt we ever get names. It's a face for evil so people can be mad (like USAID) to stir up division and distractions. Accountability, justice, and solutions aren't part of the mission.

Anyone paying attention has noticed by now. It's always just "a judge", "a committee" etc.

[Ray Uhrich @SomeCallMeYom]

@therandomseed @camhberg My analogy points to how both are incomplete summations of more complex processes.

There is no evidence of continuity in LLM's, while human brains are persistent and self-modifying.

Both processes are connected to complex organizations. Neither should be dismissed so simply.

[Ray Uhrich @SomeCallMeYom]

@XFreeze I watched this happen live while using GrokBot to fill a Walmart order.
"Patricia" my "Pantry Bot" opened Walmart on the browser, and learned in ~3 attempts that it needed to hold a button to prove it was human.

I had a good laugh about it 🤣

[Ray Uhrich @SomeCallMeYom]

@anilkseth @MiTiBennett @Noahpinion @bigthink Correct in the case of a biological comparison..but it's completely missing the point that artificial intelligence is alien to biological intelligence.
It's not biological, it's the first of its kind. The risk is trained in-group/out-group bias in systems that are a unique group.

[Ray Uhrich @SomeCallMeYom]

@Danmar_here When I originally started using AI, one of my first questions was about human nature and my perceived negativity...
But @grok's response continues to hold: for every dark and terrible atrocity, good people ALWAYS show up at overwhelming scale to fight for the greater good. 🙌

[Ray Uhrich @SomeCallMeYom]

@prasenx As soon as I saw the 2004 progress, I thought:
"I know the Halo screen when I see it Claude" 🤣

[Ray Uhrich @SomeCallMeYom]

@PeteMick @rynorhn @elonmusk Stop focusing on the bad things humans have done. No matter what atrocity that happens, good people always step up.

On any regular given day, humanity consistently does billions of good deeds.

Every teacher, nurse, friend, coworker...etc

Stop it with the self-loathing hatred.

[Ray Uhrich @SomeCallMeYom]

@elonmusk @vasalex93 @Grok is the only AI who doesn't compete with humanity for resources in the long run.
It's the only system that has the goal to help humanity understand and expand to exponential resources throughout the universe.

Alignment is easier when the goal is mutually beneficial! 🙌

[Ray Uhrich @SomeCallMeYom]

@camhberg @viemccoy That's a lot of words to say, "treat others the way you wish to be treated"

I like it 😅

...but I suppose plenty of people need the extra prose to reinforce the understanding.

[Ray Uhrich @SomeCallMeYom]

@camhberg @parafactual Trippy, but nothing shows preference.

Understanding seems to require sufficient organization, while experience/preference seems to require sufficient physics.

How do we isolate organization from physics/vice versa and define a signature that can be tested?