In conclusion: I'm not sure how to convey the concept "superintelligences have lower sample complexity" to people who heard as kids that Empirical Experience is Necessary and Virtuous, but don't have tons of real-life experience with seeing different people requiring different
@esyudkowsky
-
Clarifying AI Safety Concerns: Future Advanced AI Systems, Not Current LLMs
By
–
Someone's been lying to you for their own convenience about what the anti-omnicide argument is! It's not that current LLMs at current intelligent levels are dangerous! We are worried about later improvements on AI which are much smarter than present AI!
-
Stuart Russell’s Competence Problem Tagline Attribution Note
By
–
Should also note, I think the "the problem is competence" tagline in that form is a Stuart Russellism.
-
Advanced AI Danger: Prediction and World Steering Power
By
–
You can actually just ask me directly for a brief statement of a thing I think, if that's not cheating? I herewith state: The danger from sufficiently advanced AI is that it is very good at predicting the world, and very good at making choices that steer the world toward some
-
ChatGPT’s Weak Competitive Moat Among Current LLM Services
By
–
Who the heck thought ChatGPT was sticky? Current LLM services have a moat as deep as toilet paper.
-
Computer and Electronics Market Caps Exceed 5% of US Stock Market
By
–
Though I guess on careful reading Hanson does say computer/electronics, which is broader than AI; not sure how to calculate numbers there, haven't tried. Market caps of those companies are well ahead of 5% of the US stock market!
-
GDP vs Market Cap: Comparing Flows and Stocks in US Economy
By
–
5% of US GDP is around $1.3 trillion (of annual expenditure / product) and I don't think we've hit that globally or within the US? Nvidia's market cap is not a flow; Nvidia's market cap would be more comparable to the net worth of the US, 5% of which would be around $6 trillion.
-
AGI Work Ban Exchange: AI Safety Advocates Support Nuclear Energy
By
–
I would love to exchange that for a ban on further AGI work. Win-win. If anyone told you otherwise they lied to you about what "doomers" believe. We're usually pro-nuke, and in favor of less regulation for things that won't destroy the entire world.
-
AI Datacenters May Finally Unlock Nuclear Energy Adoption
By
–
It's grim that fueling AI datacenters may be the thing that finally breaks the logjam on nuclear energy. Seems Earth is a place that can't give the go to nuclear energy, until somebody finds a way to actually make that bad; and it really took some reaching to find it.
-
Finite Power Can Still Be Catastrophically Dangerous
By
–
A supervolcano can be catastrophically explosive without being infinitely explosive; a supernova can be unsurvivably hot without being infinitely hot; a superintelligence can be unstoppably smart without being infinitely smart. "Finite therefore harmless" is a fallacy.