Read papers. Get AI summaries.
Cite sources. Organize notes.

PaperLens sits beside you while you read — explaining every figure, grounding every answer in a citation you can check, and turning your highlights into notes you’ll actually find again.

No credit card required · Free plan includes 3 papers

Printed research charts and notes being reviewed on a desk

Summarized in 12s3 grounded citations found

Everything a careful reader needs

AI summaries

Dense sections become plain-language summaries — abstract to appendix — tuned to your reading level and the question you’re chasing.

Accurate citations

Every answer links back to the exact passage it came from, so you can verify a claim in one click before you trust it.

Organized notes

Highlights become linked notes grouped by paper, project, and topic — searchable forever, never scattered across five apps.

PaperLens

Interactive demo

Interactive Paper Demo

Lewis et al., 2020 — “Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks”

FIG. 2 · NATURAL QUESTIONS

Fig. 2 — End-to-end answer accuracy on Natural Questions as the number of retrieved passages k increases. The dashed baseline answers from parameters alone.

AI Explanation

These curves compare two ways of answering a question: reading straight from model memory (amber) versus retrieving supporting passages first and generating with them in view (indigo).

The RAG-Sequence model conditions every generated token on retrieved evidence, so knowledge-heavy questions get grounded answers instead of memorized approximations.

Accuracy climbs steeply up to k ≈ 5, then plateaus — retrieval keeps finding new evidence, but the generator’s fusion capacity saturates.
The closed-book baseline stays flat near 35 EM: the entire gap is explained by retrieved evidence, not parameter count.
Headline numbers — 44.5 EM for RAG-Sequence at k=5 vs 34.5 EM closed-book on Natural Questions.

Ask a question about this paper

Linked notes (preview)

Preview

Create a free account to build your own linked notes.

PaperLens

Plans & account

Choose your plan

Start free — upgrade when your library grows. You can switch anytime.

Free

$0/mo

For trying the full reading workflow

Up to 3 papers in your library
20 AI questions per month
Linked notes on every paper
Community support