Ryan Sun (@sun_hanchi) 's Twitter Profile
Ryan Sun

@sun_hanchi

Large Language Mystificator🧐 | Member of Non-Technical Staff @ Lehigh | Converting to JEPAism🙏

ID: 1543864209322156032

calendar_today04-07-2022 07:48:24

1,1K Tweet

239 Followers

426 Following

Ryan Sun (@sun_hanchi) 's Twitter Profile Photo

I stopped using Microsoft Office bundle: - slides --> LaTeX Beamer - docs --> Markdown - excel --> csv/json + python (matplotlib) The three replacements are text only, so I can use LLM Agents to work on them Claude Code + Cursor is my new UI for everything

Ryan Sun (@sun_hanchi) 's Twitter Profile Photo

Kaiming once used a similar analogy: one trust turbojets not because we figured out aerodynamics or solved Navier-Stokes equations, but we tested the turbojets tens of millions of times

Ryan Sun (@sun_hanchi) 's Twitter Profile Photo

Does MuP just work for MoE? I suspect the top-k operation increases variance, so maybe extreme value theorem shall be considered Maybe shift init further by 1/log, 1/loglog, or pi^2/6

Ryan Sun (@sun_hanchi) 's Twitter Profile Photo

TL;DR: Qwen3 Qwen3 Qwen3 Qwen3 Qwen3 Qwen3 Qwen3 Qwen3 Qwen3 Qwen3 Qwen3 Qwen3 Qwen3 Qwen3 Qwen3 Qwen3 Qwen3 Qwen3 Qwen3 Qwen3 Qwen3 Qwen3 Qwen3 Qwen3… (*38)

Ryan Sun (@sun_hanchi) 's Twitter Profile Photo

A reminder that 11 months have passed, and we still have no open-source implementation of o1-pro A multi-agent long reasoning framework that can use 100x test time compute to produce well-thought results

Ryan Sun (@sun_hanchi) 's Twitter Profile Photo

It’s a good mental model to think model training cost is essentially 0 It gets amortized by the increasing demand One should only care about inference cost in the long run

Ryan Sun (@sun_hanchi) 's Twitter Profile Photo

❌ Conference shitty peer review ✅ Decentralized public voting system, (e.g., huggingface daily🤗) ✅✅✅ Recommendation system for papers based on engagements I thought we solved decentralized review a while ago in social media