A 150M param recurrent model scores 29.5% on ARC-AGI-1 at $0.0007 per task
英文标题：A 150M param recurrent model scores 29.5% on ARC-AGI-1 at $0.0007 per task
Not a transformer. It's a recurrent latent reasoning setup that keeps "thinking" in latent space before answering. Sits completely outside the published cost/accuracy frontier for ARC-AGI, and something this size runs on basically anything. Paper is from the Pathway team, dropped 4 days ago. I
打开原文相关 2 · 机会 0
中文
Not a transformer. It's a recurrent latent reasoning setup that keeps "thinking" in latent space before answering. Sits completely outside the published cost/accuracy frontier for ARC-AGI, and something this size runs on basically anything. Paper is from the Pathway team, dropped 4 days ago. I
中文全文翻译正在补齐，可先查看下方英文原文。
English / 原文
Not a transformer. It's a recurrent latent reasoning setup that keeps "thinking" in latent space before answering. Sits completely outside the published cost/accuracy frontier for ARC-AGI, and something this size runs on basically anything. Paper is from the Pathway team, dropped 4 days ago. I