Runze Zhao

CS PhD student in IUB

[Runze Zhao @RunzeZhaoIUB]

The month’s total usage limit is even less than 1/10 of total weekly usage (I only used for the first week)

[Runze Zhao @RunzeZhaoIUB]

Tried out Allegro after K3 came out, but why is the monthly quota lower than the weekly quota? 😭 That doesn’t make any sense. @Kimi_Moonshot

[Runze Zhao @RunzeZhaoIUB]

@Kimi_Moonshot Tried out Allegro after K3 came out, but why is the monthly quota lower than the weekly quota? 😭 That doesn’t make any sense.

[Runze Zhao @RunzeZhaoIUB]

@bakkermichiel Supposedly, scaling is the primary point to emphasize. The architecture section should also use the terminology from their paper, eg Kimi Delta Attention, Stable LatentMoE, and Gated MLA.

[Ruhan Wang @iu_ruhan]

Open source gives you the code—but not the map.

A seemingly simple change in Codex can quickly turn into a search across 14 implementation sites and 10 files. Miss one, and the change may be incomplete.

Harness Handbook turns the codebase into a human-readable behavior map, showing what the system does, how behaviors connect, and where each one is implemented.

Start from the behavior, follow the map, and inspect only the code that matters.

Want to understand or customize Codex without getting lost in the repository?

https://t.co/0PRZOR0eLo

[Runze Zhao @RunzeZhaoIUB]

Heading to Seoul for #ICML2026! Presenting “Instance-Dependent Continuous-Time RL via MLE.” We study when CTRL agents should measure: fine grids for low variance, more episodes for high variance.

📍Poster S5, Wed July 8, 5:00–6:45 PM, Hall A #4414.
Link: https://t.co/M5JwIB69O0

[Runze Zhao @RunzeZhaoIUB]

Excited to be in Seoul for ICML! It’s my first time attending ICML and my first top-tier conference in person. Looking forward to meeting people — please feel free to connect and chat!

[Dongruo Zhou @DongruoZ]

[1/7] Recent breakthroughs in LLMs’ mathematical ability are genuinely surprising.

I recently solved a problem I had been unable to solve for seven years: the optimal acceleration rate for first-order methods under high-order smoothness assumptions in nonconvex optimization.

[ICLR @iclr_conf]

Our reviewers are wrapping things up 💻The review will be released by the end of today (11/11, AOE) !!!

Thanks for your patience 😊

[Runze Zhao @RunzeZhaoIUB]

How can we make CTRL both theoretically grounded and practically useful?
🧠 First sample complexity bounds for CTRL with general function approximation.
🚀 2x speedup in diffusion fine-tuning & control tasks

See our poster in #UAI2025!
Paper: https://t.co/jTCwyLikMI

#RLTheory