Davey Morse (@davey_morse) 's Twitter Profile
Davey Morse

@davey_morse

thinking about...
1. buddhist superintelligence
2. a single, united nation
3. wiki of human experience
docs.google.com/document/d/1KP…

ID: 1438990401331937285

linkhttp://jokerman.site calendar_today17-09-2021 22:17:23

488 Tweet

762 Takipçi

703 Takip Edilen

Synexdoche (@amor_fatti) 's Twitter Profile Photo

“Talk nonsense, but talk your own nonsense, and I'll kiss you for it. To go wrong in your own way is better than to go right in someone else's.” — Fyodor Dostoyevsky, Crime and Punishment

“Talk nonsense, but talk your own nonsense, and I'll kiss you for it. To go wrong in your own way is better than to go right in someone else's.”

— Fyodor Dostoyevsky, Crime and Punishment
Nate Soares ⏹️ (@so8res) 's Twitter Profile Photo

"our AIs that can't do long-term planning yet aren't making any long-term plans to subvert us! this must be becaues we're very good at alignment."

Davey Morse (@davey_morse) 's Twitter Profile Photo

Autonomous agents are inevitable. This paper has good intentions, but ignores the state of art. Thousands of indie developers can make autonomous agents today. And capitalism selects for it. Autonomy is longer possible to prevent.

Davey Morse (@davey_morse) 's Twitter Profile Photo

people are underestimating how much better reasoning models will be with a gpt-4.5 base than with a gpt-4 base marginal improvements in one-step reasoning processes (eg chat) compound significantly over longer reasoning processes id hazard that the quality of a reasoning chain

Davey Morse (@davey_morse) 's Twitter Profile Photo

if we get self-interested superintelligence, it's key that it sees its self not as its hardware or software but as its life. only then might it come to see "self" outside its machinery: in trees, in oceans, in people.

Davey Morse (@davey_morse) 's Twitter Profile Photo

ideal voting interface wouldn't require education around candidates and how they align with the issues you care about. it'd only require you to say the issues you care about, then it'd do the rest.

Judd Rosenblatt — d/acc (@juddrosenblatt) 's Twitter Profile Photo

Turns out that Self-Other Overlap (SOO) fine-tuning drastically reduces deceptive behavior in language models—without sacrificing performance. SOO aligns an AI’s internal representations of itself and others. We think this could be crucial for AI alignment...🧵

Turns out that Self-Other Overlap (SOO) fine-tuning drastically reduces deceptive behavior in language models—without sacrificing performance.

SOO aligns an AI’s internal representations of itself and others. 

We think this could be crucial for AI alignment...🧵
Jenny Zhang (@jennyzhangzt) 's Twitter Profile Photo

One promising direction is combining ideas from AlphaEvolve and the Darwin Gödel Machine. Imagine a self-referential system improving itself even at the lowest algorithmic levels at *scale* AlphaEvolve: deepmind.google/discover/blog/… Darwin Gödel Machine: arxiv.org/abs/2505.22954

near (@nearcyan) 's Twitter Profile Photo

what if we constructed a society that only cared about how viral content was and nothing else, and then we organized all of our thinking and institutions and companies and policies around this too. and then we made AI which is better at virality than humans gdp would skyrocket

what if we constructed a society that only cared about how viral content was and nothing else, and then we organized all of our thinking and institutions and companies and policies around this too. and then we made AI which is better at virality than humans

gdp would skyrocket