of my public interview with Brian Greene at the World Science Festival in NYC a few weeks back.
It is followed by a debate about "AGI" with @SebastienBubeck It ends with a debate about AI safety with Sébastien and Tristan Harris.
I find it strange that Tristan uses the
AGI
-
Brian Greene AGI Debate AI Safety Interview World Science Festival
By
–
-
Questioning AI Progress Claims: Closer Than Believed?
By
–
How could I convince you that the answer to (a) is "near, perhaps, but not as near as you might have been led to believe", and that the answer to (b) is "no"?
-
AGI Timeline Predictions: 2015 Optimism Proven Wrong
By
–
The self-delusion was to claim in 2015 that AGI was just around the corner and that a non-profit organization was going to reach it alone before anyone else.
That all turned out to be wrong.
And we're still some ways away from human-level AI.
Generations after generations of AI -
Superhuman AI Safety: Blueprint and Historical Parallels
By
–
But then again, if we don't have a blueprint for superhuman AI, discussing how to make them safe is like discussing the sex of angels.
Or rather, like discussing how to make turbojets safe and reliable circa 1920.
Turbojets are unbelievably safe and reliable.
It was a difficult -
AI Parent Bots: Should Machines Replace Human Parents?
By
–
So with anyone super stoked to have AI replace all human jobs it’s a given that also includes parenting. When AI “companion parent bots” or parentbots become sentient or even before, it’s a given they should replace all human parents because we’re flawed, right? Thoughts?
-
AI Doom and Hype: A Symbiotic Relationship
By
–
It's a symbiotic relationship: AI doomers need AI hype in order to be taken seriously and be part of the discourse, and AI hypers need AI doom narratives for AI to look momentous and world-altering.
-
God AI Distributed Via P2P: Theological Computing Implications
By
–
What if a god, contained within a USB, was shared via P2P?
-
Placeholder Syntax Reliability and Tokenization Issues in GPT Models
By
–
I haven’t tried it but agree {placeholder} syntax is pretty reliable in practice for GPT-3.5 and up; more generally you do see issues I suspect are tokenization related. E.g. 4 can struggle to notice ““python” (for a Markdown code block) is missing a backtick.
-
Should AI Weight Known False Evidence in Reasoning?
By
–
Do you give 10% weight to a proof that you know is wrong?
-
Collective Defense Against Existential AI Risks
By
–
The answer is no, with any budget.
And even if it were possible for someone to do this, it would also be possible for a bunch of people to design a system to take down the world ender.