切问学术

Overview

切问学术 is a research-focused AI agent developed by the NLP team at Fudan UniveRSIty. Positioned as an academic counterpart to Coding Agents, it AIms to reduce the repetitive overhead of scientific work so researchers can spend more time on ideas. The platform combines literature analysis with hands-on experiment support, covering a workflow that usually spans several separate tools and manual steps. Because the product is continually evolving, specific capabilities and availability may vary.

What it does

The Agent is built around two main jobs: turning a paper into a runnable experiment, and helping researchers sharpen a research idea. On the experiment side, a user uploads a paper PDF, and the system estimates the complexity of the described experiments and the compute they require. It then prepares a GPU server and the matching software environment, such as PyTorch and CUDA, and proceeds toward reproducing the work.

Key capabilities

• Automatic paper reproduction: converts a submitted paper into a configured experiment environment with minimal manual setup.
• Self-directed troubleshooting: when dependency conflicts or script errors APPear, the agent can try alternate installations or rewritten scripts until the run completes.
• Visible progress: users can watch the process in real time, from environment construction through experiment execution, similar to a live stream.
• Idea discovery: a mentor-style dialogue helps users organize scattered thoughts into a structured research direction.
• Literature analysis: deep retrieval of related work highlights gaps and limitations, then assesses the novelty of an idea and how it differs from existing studies.
• Long-task handling: the agent can process large reading loads, for example summarizing roughly 200 papers into a lengthy review document.

Who it is for

The tool suits graduate students, postdoctoral researchers and faculty who need to validate prior results or explore new directions. It is especially relevant for teams limited by compute access or configuration expertise, since the agent handles environment setup and common failures. Research groups that read extensively before committing to a topic may also benefit from the literature synthesis and novelty assessment features.

How to get started

Access is through the web platform. A typical first step is submitting a paper PDF and reviewing the agent's estimate of experiment scope, followed by observing the automated setup and run. For idea work, users start a dialogue and supply their area of interest. Chinese-language support is central to the experience, so users outside that audience may find the interface less convenient.

Pricing and access

Pricing details are not clearly stated and may change as the service matures, so check the official site for current terms. Compute-heavy features such as full reproduction could involve costs or usage limits that may vary by plan. Treat any free tier as subject to change and confirm access conditions before relying on the tool for time-sensitive research.