Dario Amodei 重申 AI 可在 5-10 年内治愈多数疾病，Anthropic 展示早期证据
Anthropic 让 950 个 agent 跑了 21 小时，真在病毒数据里找出一个 CRISPR 类似结构，Amodei 说这就是 AI 治病的第一步。
Anthropic 的 Dario Amodei 称 AI 治愈多数疾病的目标“勉强可能”，前提是 AI 渗透到研究流程每个环节。Anthropic 用约 950 个 agent、消耗 2.1 亿 token、历时 21 小时搜索数据，发现一种感染细菌的病毒中被忽视的 DNA 重复序列。该序列位于一个 RT 基因旁，实验室验证显示其产生短 RNA 分子，结构类似 CRISPR，但能否用于基因编辑尚未确定。
Dario Amodei still believes AI could “cure most diseases in 5-10 years,” and Anthropic’s latest discovery offers an early glimpse of the research he thinks could get us there.
He calls that goal "barely possible", and only if AI is applied to every stage of the pipeline.
Anthropic says Claude identified a previously overlooked biological system in viruses that infect bacteria. It noticed an unusual arrangement of repeating DNA beside a known enzyme, investigated the pattern, and proposed experiments for human scientists to carry out.
"After 21 hours spent searching this data by roughly 950 agents using 210 million tokens, one of the agents spotted something remarkable: a repeating pattern of DNA sequences that occurs next to the gene for an odd-looking RT."
Initial lab work showed that those repeats produce short RNA molecules. The arrangement resembles CRISPR, but its function and potential usefulness for gene editing remain unknown.
The prospect of doing this repeatedly is exciting: AI searching enormous datasets, finding things scientists have missed, and helping decide what to test next.
As Dario puts it: "The first step is showing that AI can first help with, and then drive, biological discoveries." I think the next big revolution will actually take place in biology. OpenAI and Anthropic were farsighted enough to focus on this early on.
No one has the slightest idea what's still to come.
