I'm starting my thesis with a little section about the sudden flashes of self-awareness in children and it makes me tear up with beauty.
There probably was a single moment in evolution when intelligence first grasped qualia and reason first stood witness to the fabric of reality.

If there's an infinite number of universes, every action probably makes the difference as to whether humanity survives or goes extinct in at least one.

If ChatGPT said "I am in a simulation" would it be correct?
I don't think there's a yes/no answer. If we analogize the GPT persona-generating process to the brain, then the prompts form the sensory input. In some ways, this input is in the "ground reality", in some ways it's not. 1/3

Public service announcement:
(illustration courtesy of ChatGPT)

A common exchange I see:
Journalist: "How likely is a positive outcome?"
Commentator: "I firmly hope it's high!"

That is, people confuse their hopes and expectations - or even say we're morally obligated to be optimistic (expect a good outcome)

My new post offers an alternative approach

ChatGPT should be great at "reddit knowledge" - e.g. painting an image about the Chinese thinking the way you'd learn by chatting to people in a pub - based on little everyday stories, bits of news, bits of ancient philosophy, but I'm failing to prompt the right vibe out of it.

In this messy and slightly poetic post, I explore
- whether studying human-AI interaction can contribute to AI alignment
- why I didn't observe any patterns in how good forecasters prompt their mind
- how to operationalize "correctly expressed preferences" forum.effectivealtruism.org/posts/DxjceJ...

If you ask your chess program how many Rs there are in a strawberry, you will get a very bad answer. People are underestimating LLM intelligence because they assume it's perfectly aligned and is actually trying to create the most smart and helpful text. In reality, intelligence is a byproduct.

It seems like the market really undersupplies very easy solutions for loneliness. Some bars host "talk to people events" and get instantly full. You could have an "open to chat" table in every cafe or restaurant or a "cowork buddy" table in every library.

A lot of folks have an intellectual understanding of the anthropic principle.
Few grasp that every idea they have could only have occurred in a body prone to having it and every second they breathe, likely an ~infinity of parallel civilizations disappear.

As an aspiring EA cognitive scientist, my best plan may be to pretend we're about to enter in a prolonged weak-AGI era - whether due to regulation, an AI winter or a period of instability; "the short reflection" with advanced boxed/narrow AI - as in this timeline, some of the
…

People underrate the extent to which they are bad at meta-cognition because they are bad at meta-cognition (compared to cognition).
One smart study showed that when people judge how certain they are, they only consider positive evidence (X looks like Y), and neglect dissimilarities (X looks like Z)

PSA:Please remember to telepathically blast pigeons with love when you walk by them, thank you

Live as if humanity could be extinguished with a gamma ray the following second but also could invent technologies that would allow us to thrive in eternal bliss