AI 研究者终于有了能并行跑假设验证的工具——FARS 把实验室一年的工作量压缩到一周，做学术或工业研究的团队可以直接用它加速探索。
Lemma 推出 FARS（全自动研究系统）模式，通过四个专门智能体（构思、规划、实验、写作）实现端到端的 AI 研究自动化。该系统无需人类干预即可运行完整研究循环，从提出假设到撰写论文。FARS 通过共享文件系统协调智能体，支持并行运行多个研究线程，使单个研究者一周内能完成传统实验室一年的工作量。这有望彻底改变当前研究效率低下的现状，解决文献综述耗时、路径选择风险高等痛点。
The way we do research is about to change. Today,…
The way we do research is about to change.
Today, doing research is overwhelming. There's so much literature to review and so little time that picking the wrong path could take you back years.
We might have a solution for this:
There's a new multi-agent system with four specialized agents to do AI research at scale:
Ideation → Planning → Experiment → Writing
Lemma's FARS (Fully Automated Research System) mode was built to produce hypothesis-validation pairs so that a single researcher can explore in a week what used to take a lab a year.
Their four agents can run a full research loop end-to-end without relying on humans at all.
Each agent owns one stage of the pipeline:
• The Ideation agent finds the hypothesis • The Planning agent designs the study • The Experiment agent runs it • The Writing agent drafts the paper
This design allows FARS to run multiple research threads in parallel, like a lab with dozens of researchers working simultaneously.
I read their blog post, and the way they architected the system is pretty interesting:
The agents don't talk to each other directly. Instead, they coordinate through a shared file system that doubles as a workspace and a persistent memory.
This allows the system to scale to many concurrent projects.
Here is what FARS gives you:
You give it a research direction → The system generates hypotheses → Runs them through automated reviews → The ones that pass move down the pipeline → You get back short papers focused on a single, well-scoped contribution, each one is a hypothesis and its validation.