AI Paper Atlas

Research map / Multi-agent systems and games / Human-AI and goal-conditioned

WebShop: Towards Scalable Real-World Web Interaction with Grounded Language Agents

Shunyu Yao, Howard Chen, John Yang, Karthik R Narasimhan

NeurIPS 2022 · Multi-agent systems and games › Human-AI and goal-conditioned · approach: World-model-based agents · 51 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
fine-tune IL agent with RLIL agent · WebShop testtask score: 59.9 -> 62.4Improved
fine-tune IL agent with RLIL agent · WebShop testsuccess rate: 29.1% -> 28.7%Hurt
RL fine-tuning makes agent greedier, lowering option scoreIL agent · WebShop testoption score: 45.2 -> 38.9Hurt
add language pre-training (BERT init) to choice model (ablation)IL agent · WebShop testsuccess rateImproved
add BART search query generator instead of instruction-text rule (ablation)IL agent · WebShop testscore and success rate: around 3 pointsImproved
add history (one past observation, last five actions) (ablation)IL agent · WebShop testtask score: 59.9 -> 57.3Hurt
warm-start RL from IL instead of directly from pre-trained BERT (ablation)RL agent · WebShop testtask scoreImproved
use Transformer (BERT) encoder instead of RNN in RL (ablation)IL+RL · WebShop testsuccess rate: more than 10% betterImproved
use top-1 search query instead of sampling from top-5 (ablation)IL agent · WebShop testscore: 60.56 -> 61.96Improved
use top-1 choice instead of sampling (ablation)IL agent · WebShop testscore: 60.56 -> 45.10Hurt
add image features (ResNet) (ablation)IL agent · WebShop testscore: 60.3 -> 60.6No clear effect
use Choice oracle with human last query instead of instruction text queryChoice oracle · WebShop testsuccess rate: 85.4% -> 87.8%Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy