Solving poker in custom WebGPU kernels

Hacker NewsPublished Jul 30, 2026

A developer used OpenAI's Codex to create custom WebGPU kernels for running a poker solver neural network directly in the browser, achieving over 10x speedup and parity with PyTorch models without existing tensor libraries.

#neural-network #openai-codex #poker-solver #pytorch #webgpu

Solving poker in custom WebGPU kernels

More in Dev

Moonshot releases Kimi K3.1 Pro and says it cut its hacking skills

Microsoft Open-Sources Agensh: A Multi-Agent Harness for 1,024 Agents

Cognition Open-Sources an Agent Plugin that Helps Secure Systems against Recent Cybersecurity Threats