Skip to content

Blog ​

The arguments behind the work: how the system is put together, why a loop is shaped the way it is, and what a long run taught us that a number alone would not. The results themselves, one per post, are in the news. Subscribe by RSS.

Flow Science

Flow Science: A New Dimension of Scaling Agents

Models scale with parameters and data, and agents scale with tools; the next dimension is the flow. Why one flow cannot solve every problem, flow-level ablations on four benchmarks with the models held fixed, four findings we did not expect, and whether a flow is engineering or a science.

Zijian ZhangRead it

KDA

KDA²: Kernel Design Agents (KDA) Optimize Kimi Delta Attention (KDA)

Dongyun Zou, Yixin Dong, Hongyi Jin, Junxian Guo, Yahui Cui, Avery Huang, Zihao Ye, Junru Shao, Changye Li, Zijian Zhang, Sihao Liu, Song Bian, Ligeng ZhuRead it

Method

RLAR: The Review Is the Next Prompt

HumanfiaRead it

Architecture

Humanfia: From a Review Loop to Four Layers and a Referee

HumanfiaRead it

KDA

A Quality Tier for the Fast Kernels

Most profitable diffusion fusions change the output at the level of bf16 rounding order, enough to fail a bitwise CI and not enough for a person to see. So we gave the request a quality tier instead of arguing about the diff.

Xiaoyu ZhangRead it

Humanize

Model, Tool, Flow: 2, 18, 50 of 50

The same model, measured three ways (raw API, the vendor's CLI, and that CLI inside a flow) on PutnamBench, Physics Cup, SuperChem and an HLE subset. Putting the CLI inside a flow adds more than the CLI's own tools do.

Zhengyang Zhang, Changye LiRead it

KDA

Lossless Is a Choice

A competing stack bought its speed with approximate kernels and skipped steps. Reading it closely turned into three PRs that take about a fifth off our LTX-2.3 time on B200, two of them byte-identical end to end.

Xiaoyu ZhangRead it