Abstract: Moloch, demon god of Inadequate Equilibria and Coordination Problems, still rules over human affairs, though his grasp has weakened significantly from the Malthusian Traps our ancestors were captured in.
From the rapidly strengthening steel of AI technology, we are promised to forge a weapon mighty enough to fight back his demonic influence more still, perhaps even slay the tyrant for good.
But how should we conduct this war? How can we win? How could we lose?
Our technology is giving us rapidly growing power to build the paradise we want. I will argue - that the instinctive models of how to use this power are fundamentally broken - that adding more people to the cause may be paradoxically a net negative (in certain circumstances) - that some organizations are using the language of AI Safety to misdirect creative energies
I'll point out some hopeful directions I see towards the alignment problem.
Bio: Connor is a founding member of the loose AI research collective EleutherAI. They are most well known for their ongoing efforts to replicate GPT3. Recently, Connor has been pushing for more directly alignment relevant research at Eleuther, and invites anyone with ML or webdev experience to join. By day, he works as an AI Researcher at German startup Aleph Alpha.
The event begins 20:30 Israel Standard Time, 10:30 Pacific Standard Time, 18:30 UTC.
Connor Leahy will speak on "Moloch Reigns: To Gather an Army or to Forge a Sword?," then will take questions.
After that, we will socialize virtually at the LessWrong Walled Garden in GatherTown.
Please register at https://forms.gle/zxrR3N763bNsAkAS9 and we will send you an invitation by January 30.
Abstract: Moloch, demon god of Inadequate Equilibria and Coordination Problems, still rules over human affairs, though his grasp has weakened significantly from the Malthusian Traps our ancestors were captured in.
From the rapidly strengthening steel of AI technology, we are promised to forge a weapon mighty enough to fight back his demonic influence more still, perhaps even slay the tyrant for good.
But how should we conduct this war? How can we win? How could we lose?
Our technology is giving us rapidly growing power to build the paradise we want. I will argue
- that the instinctive models of how to use this power are fundamentally broken
- that adding more people to the cause may be paradoxically a net negative (in certain circumstances)
- that some organizations are using the language of AI Safety to misdirect creative energies
I'll point out some hopeful directions I see towards the alignment problem.
Bio: Connor is a founding member of the loose AI research collective EleutherAI. They are most well known for their ongoing efforts to replicate GPT3. Recently, Connor has been pushing for more directly alignment relevant research at Eleuther, and invites anyone with ML or webdev experience to join. By day, he works as an AI Researcher at German startup Aleph Alpha.
The event begins 20:30 Israel Standard Time, 10:30 Pacific Standard Time, 18:30 UTC.
Posted on: