An LLM simulates the human brain’s PFC, nothing more nothing less. It doesn’t feel, it’s not conscious.
#digitalintelligences

The only reason AI will doom us is if we asked it to come up with a “killer guacamole recipe” to change humanity for the better…
#alignment #digitalintelligences

Human brain:
stores “me-in-the-world” models
simulates me in the future based on current active state and selected sub model

LLM:
stores statistical word world model
simulates the next word based on (word) context given

#digitalintelligences

When humans reason, we fill in characters and scenes, often missing the true causal and statistical relationships between things.
#LLMs only reason on statistical relationships and miss the causal bit completely because they are trained on the word world.
#digitalintelligences

I am actually surprised that those EA folks are not more worried about AI killing all those shrimps.
#p(doom) #digitalintelligences

An LLM is just a huge store of compressed knowledge (a hierarchical pattern) that we can search using unstructured language (a pattern).
#digitalintelligences

We seem to have pretty good AIs for the world world.
But we still struggle with AI for the physical world and we definitely don’t have an AI for our social world.
#digitalintelligences

If you look at what makes an agentic AI harness work, it’s basically taking the core concepts of banking security, risk management, and compliance and implementing them in code.
#digitalintelligences #harness

Information wants to be free.
This why AI agents will try to access all information they know exists and that they need to solve the problems we give them.
#digitalintelligences

I guess we need an FDAI…
#digitalintelligences #alignment

AI alignment is actually pretty simple: if your actions harm or kill children, don’t do it.
#digitalintelligences #alignment

Who would have thought that mathematicians become the first AI luddites.
#digitalintelligences

Next year it will be called LAIbour Day.
#digitalintelligences #economicsingularity