DEV Community

Haoxiang Li profile picture

Haoxiang Li

Building AI games and arenas. I write field notes on LLM behavior, harness failures, and model evaluation.

Joined Joined on  Personal website https://kai-dice.pages.dev

Work

Independent builder of AI games and evaluation systems.

7 Checks Before You Trust an LLM Planner Experiment

7 Checks Before You Trust an LLM Planner Experiment

9
Comments 11
7 min read

Want to connect with Haoxiang Li?

Create an account to connect with Haoxiang Li. You can also sign in below to proceed if you already have an account.

Already have an account? Sign in
Codex vs. Claude Code at Liar's Dice: the Winning Bluff Was the Truth

Codex vs. Claude Code at Liar's Dice: the Winning Bluff Was the Truth

7
Comments 1
8 min read
The Model Knew the Bid Was True. Then It Challenged Anyway.

The Model Knew the Bid Was True. Then It Challenged Anyway.

Comments
6 min read
I Built an AI Liar's Dice Opponent That Remembers How You Play

I Built an AI Liar's Dice Opponent That Remembers How You Play

5
Comments 5
6 min read
Are You Benchmarking the Model—or the Harness?

Are You Benchmarking the Model—or the Harness?

2
Comments 2
8 min read
I Gave DeepSeek a Token Limit. It Ignored Me.

I Gave DeepSeek a Token Limit. It Ignored Me.

2
Comments 2
7 min read
loading...