Excited to share our new work on Automatic Wikipedia Updating with LLM Agents!
We introduce WiNELL, an agentic framework that continuously monitors online sources for recent facts, identifies relevant updates for the Wiki article under consideration, and generates well-formed edit suggestions.
Paper: github.com/gangiswag/Auto…
We introduce WiNELL, an agentic framework that continuously monitors online sources for recent facts, identifies relevant updates for the Wiki article under consideration, and generates well-formed edit suggestions.
Paper: github.com/gangiswag/Auto…
Wikipedia's updating process is manual, and there can be a considerable latency between when the event happens vs when it propagates to Wikipedia.
Even popular pages have update latencies ranging from weeks to months.
Even popular pages have update latencies ranging from weeks to months.
WiNELL leverages multi-agent search to identify relevant online updates about a given article.
A fine-grained editing model (trained on human revision histories) then incorporates the update into the appropriate Wikipedia section.
A fine-grained editing model (trained on human revision histories) then incorporates the update into the appropriate Wikipedia section.
Our end-to-end automatic evaluation is based on factual coverage by measuring the extent to which agent updates entail the atomic facts in historical human edits.
Here is a qualitative example for @LewisHamilton's Wikipedia page, where the agent identified and extracted information from multiple news sources, to generate its update for the correct subsection (@ScuderiaFerrari 2025), successfully capturing 3 out of 4 atomic facts (in green).
Work done in collaboration between UIUC and Amazon @tanayd07 @qiancheng1231 @daniel_js_lee @Ruhi_Sarikaya @hengjinlp
• • •
Missing some Tweet in this thread? You can try to
force a refresh
