Our cloud community, so far. Just the beginning.

RT @SchmidhuberAI: Everybody talks about recursive self-improvement & Gödel Machines now & how this will lead to AGI. What a change from 15….

RT @DefenderOfBasic: all it takes is one mystic-led lab to make a scientific breakthrough and then everyone will be scrambling to court a g….

Best slide of the year from @psychosort talk today at Network School

RT @hardmaru: Had a really fun panel discussion with @jeffclune and other heroic minds at the World Models workshop at #ICLR2025 https://t.….

RT @zhengyaojiang: Tim’s @_rockt keynote at ICLR showcased @WecoAI, a small startup in London that is playing in the grand game of AI-drive….

RT @hardmaru: Tim Rocktäschel’s keynote talk at #ICLR2025 about Open-Endedness and AI. “Almost no prerequisite to any major invention was….

Update on the state of ML/AI research in 2025 for those who think AI has become commodity. >10,000 in attendance, not an LLM wrapper in sight. #ICLR2025

RT @jeffclune: Very excited for this keynote by @_rockt! Awesome to see open-endedness go from a niche (😉) area to a keynote at #ICLR ! 🌱🌿🌳….

RT @hardmaru: We should host more top ML conferences (ICLR, ICML, NeurIPS) in Asia.

RT @S_Conradi: 510 million white dots. Each dot is a root of: 𝑥¹⁷ + (- 20i t₂⁷ + 20i t₂⁶ - 20i t₂⁵ + 20t₂⁴ - 20t₂³ + 20t₂² + 20t₂ - 20i)𝑥¹⁰….

RT @RichardSSutton: Rich's slogans for AI research (revised 2006):.1. Approximate the solution, not the problem (no special cases).2. Drive….

RT @balajis: Anyone. Anywhere. $100k in funding. Global meritocracy is finally here.

RT @krrishd: “content” is such a viscerally impotent word because it’s never the actual thing anyone wants.

Yes, I believe Δiq is a thing.

Just to elaborate on the preference tuning bit:. - Assemble a community of people willing to closely follow the advice of their LLM. - Collect objective measures of success daily from people (Δiq, Δhrv, Δ$). - Use this as the sparse reward for preference tuning.

As my LLM gets to know me, it should be more brutally honest day by day. Not more sycophantic, that's just annoying. Point out my blind spots, help me identify the one thing that is holding me back. I think dedicated preference tuning is needed here rather than prompting.

RT @DefenderOfBasic: it was theorized 45 years ago, "the telescope looks at the infinitely great, the microscope looks at the infinitely sm….

Moments in life experienced as flow, mastery, or enlightenment occur when this fractal nature reveals itself explicitly. True examination of gross reality reveals subtle reality. And the gross:subtle relation is universal. All roads lead to the top of the mountain.

This is also true of cognition generally. The goal in learning should be to capture truths in a fractal way, such that new truths add resolution to your model without the need for radical overhaul of the existing structure.