I increasingly believe that there are fundamental principles which simultaneously govern the designs of well-functioning minds, organizations and societies. Once we pin them down with mathematical precision, we’ll understand the world more deeply than we can currently imagine.

RT @DavidDuvenaud: The second keynote was @RichardMCNgo on “Flourishing in a highly unequal world”. He argued that future beings will vary….

Blazing-fast image creation – using just your voice. Try Grok Imagine.

RT @HiFromMichaelV: It was also a way to create a permanent state of exception and preserve the military mobilization of the Second World W….

RT @MatharyCharles: Today's reading:. In effect, re-interprets the outer optimization step in DiLoCo / FedOpt as a form of batched model en….

RT @darengb: @nickcammarata On the Origin of Species actually cited Empedocles as the first source for a theory of natural selection in in….

The very best researchers instead do step 2: “understand the formalism and its limitations”. But most instead do step 2: “accept the formalism and work within it in a way that subtly constrains their thinking”. See e.g. ML researchers confronting the rise of deep learning.

I’ve often gone through a cycle of:.1. Try to understand a formalism.2. Bounce off it.3. Think “well I’ll just try to understand the phenomenon on an intuitive level for now”.4. Come back later and realize that the formalism is actually very limited and badly-motivated. Examples:.

Sometimes I wonder whether my biggest comparative advantage as a researcher is that I’m not good enough at math to get lost in formalisms.

RT @Tim_Hua_: AIs can have secret “backdoors:” Can we uncover them?. Maybe! We study models that misbehave in an unknown situation, and suc….

Only a few minutes after posting this, an example popped up in one of my group chats. A leftist is claiming that the right runs the country now, and a rightist is responding by arguing that the left still runs all the important institutions. Underdog bias is everywhere.

New post on underdog bias, which I think is the most important cognitive bias for understanding modern society.

RT @tracewoodgrains: Unidentifiable, unaccountable officers are an exceptionally bad idea. How is this sort of thing distinguishable to onl….

I keep seeing people and movements cause the opposite of what they were trying to achieve. This phenomenon seems important to understand if you want to actually move towards your goals. I call it pessimization, and I write about it in my latest post (link below).

RT @joannejang: i think everyone in ai should think about what their "line" is. where if your company knowingly crosses that line and won'….

RT @james_s48: An approximation I like is that an H100 is roughly a studio apartment

@nathancofnas Another recent thread on this broad idea:.

@HiFromMichaelV @ben_r_hoffman @benlandautaylor Since most people won’t read the posts, and the thread is kinda abstract otherwise, here are the two most relevant excerpts. This thread is also very relevant:

Consenses of power and centralized conspiracies are both allergic to transparency. But consenses of power are more like a disease to diagnose than a conflict of fundamental interests. Thanks to @HiFromMichaelV, @ben_r_hoffman and @benlandautaylor for helping me understand this.

I call these patterns “consenses of power”. They’re emergent/distributed conspiracies which, despite some superficial similarities to centralized conspiracies, have very different dynamics. More in this post (and the follow-up, Elite Coordination via the Consensus of Power):.