YC Paper Club explores AI compute beyond GPUs
What happened
On October 3, Hacker News featured a YC Paper Club session on compute alternatives to GPUs, covering optical computing, neuromorphic architectures and biological computing. Topics included methods that go beyond backpropagation, zeroth-order optimization and SOMA, building diffusion models with light, and training biological neurons with reinforcement learning to play Doom. The discussion suggests AI hardware may diverge from its tight coupling with transformers, backpropagation and GPUs as compute efficiency plateaus. The report covers these alternative routes only; no deployment progress is disclosed.
Written by AI from the coverage · updated 1 hour ago
Coverage
Follow the reports to see the story from different sides.
- Hacker News front pageWhat if we stopped using GPUs? [video]
YC Paper Club 本期聚焦 GPU 之外的替代算力，探讨光学计算、神经形态架构与生物计算三条路线。内容涵盖超越反向传播、零阶优化与 SOMA、用光构建扩散模型，以及用强化学习训练生物神经元玩 Doom。讨论认为，随着算力效率开始进入平台期，AI 硬件可能出现与 transformers、反向传播和 GPU 紧耦合截然不同的形态。
Heat over time
Not enough continuous observations to draw a trend yet.