← machinelearningpapers.com

Agent Skills

How to use this site with AI agents (Claude, ChatGPT, etc.) for ML research

What This Site Is

machinelearningpapers.com is a continuously updated database of ML papers fetched daily from arXiv, HuggingFace, and Reddit — analyzed by LLM for relevance and impact. Each paper gets a structured analysis: relevance reasoning, methodology assessment, limitations, broader impact, and an impact score (0–100).

The Best of All Time page is a hand-curated collection of landmark papers with significance notes. The main index shows papers from the last 14 days with impact score ≥ 84.

audio vision llm interpretability general_ml robotics systems bio

Using With an AI Agent

The easiest way to use this site with Claude or another agent is to share a page URL and ask it to fetch the content. Every page is plain HTML — the agent can read paper titles, abstracts, scores, and analysis fields directly.

Pages you can share:

Example Prompts

Literature review before starting a project

Fetch https://machinelearningpapers.com/best.html and https://machinelearningpapers.com and give me an overview of the current state of the art in [your area]. Focus on papers in the [audio/vision/llm/etc] category.

Staying current with the field

Fetch https://machinelearningpapers.com and summarize the most impactful papers from the past two weeks. Which ones are most relevant to [your problem]?

Finding prior work on a specific problem

Fetch https://machinelearningpapers.com/best.html and search for papers related to [your problem: e.g. "training stability", "audio generation", "mechanistic interpretability"]. What approaches have been tried and what are their limitations?

Understanding a paper's significance

Fetch https://machinelearningpapers.com/best.html — find [paper name or arXiv ID] and explain its contribution, methodology, and what papers it influenced.

Debugging a training issue

Fetch https://machinelearningpapers.com/best.html and https://machinelearningpapers.com — find papers about [e.g. "catastrophic forgetting", "training instability", "reward hacking"]. What solutions have been published and what are the tradeoffs?

Impact Score

Each paper is scored 0–100 from three components:

Score ≥ 84 → main feed (roughly 1–5 papers/day). Landmark papers bypass thresholds entirely.

Browsing the Site Directly