Industry9月22日 21:28

WebArena作者Shuyan Zhou入职Meta超级智能实验室

量子位@衡宇

Blog/Media

Read the original

Why we picked it

WebArena作者Shuyan Zhou加入Meta超级智能实验室，他的梦想是开发能帮妈妈订机票的AI浏览器。

WebArena作者Shuyan Zhou已加入Meta超级智能实验室。WebArena是一个用于测试AI系统在真实网页环境中表现的基准测试平台。

Image · 量子位

Body

Shuyan Zhou曾致力于开发能帮助用户完成复杂网页任务的AI系统。他的目标是创造能够帮助普通人完成日常任务的AI浏览器。

Original · 量子位

梦想是做能帮妈妈订机票的AI浏览器。

Industry Coordinates & Source Verification

This report is categorized under AITOP's "Industry" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Key entities and technical topics under continuous tracking include WebArena, Shuyan Zhou, Meta, AI browser.

Source attribution: curated from 量子位 (@衡宇).

#WebArena #Shuyan Zhou #Meta #AI browser

Keep reading

Recent stories with the same tag or section

Papers Builders14:01

Chinese contentJHiMA-WebRL：分层多智能体强化学习框架提升长程网页任务表现

用两个 7B 模型分工干翻 32B 单体：一个负责拆任务，一个负责执行，在 WebArena-Lite-v2 上成功率 37.2% 对 24%，做 GUI Agent 的可以看看训练细节。

#JHiMA-WebRL #WebArena #GRPO #GUI Agent

arXiv cs.AI@Shanka Subhra Mondal, Jiayun WangSource

Papers Builders14:59

Chinese contentSCOPE 框架：让 LLM Web 智能体从局部技能学成长程任务能力

训练 Web 智能体老是拿不到反馈？这篇把复杂任务拆成技能链一步步学的思路挺有意思，三个基准上最多提升 18.6%。

#WebAgent #SCOPE #WebArena #agents

arXiv cs.AI@Xueyang Feng et al. (5)Source

Papers BuildersSelected00:11

SFResearch Study Tests Fragility of Memory-Based Self-Improving Agents

The takeaway is that agent self-improvement may be more fragile than reported scores suggest, because evaluation conditions differ sharply from production task streams. Teams building memory-based agents should run multi-seed evaluations, randomise task order, and enrich memory construction with explicit rubrics and feedback. Code and data are released openly, so practitioners can replicate the findings directly.

#agents #WebArena #memory #agent evaluation

Official accountSalesforce AI Research@SFResearchSource

Papers BuildersMulti-sourceSelected10:03

Salesforce Research's CLIFT Lifts 31B Gemma-4 Agent Past Gemini 3 Flash

The key implication is that a mid-size open model can beat a frontier model with browser use on web-agent benchmarks by replacing an expensive judge with a self-verification question bank. Teams building browser agents can adopt CLIFT to cut judge API calls at both training and deployment. Because it starts from Gemma-4, it offers a concrete open-source alternative to proprietary agents like Gemini 3 Flash.

#Salesforce #Gemma-4 #Gemini 3 Flash #agents

Event

DAIR.AI@dair_aiSource

Papers Builders13:25

Chinese contentCLIFT：用保形自验证训练开源网页智能体并替代外部裁判

开源网页智能体训练时奖励信号太稀疏？这篇 CLIFT 用保形自验证把裁判反馈变成可复用信号，测试时还能不调 API 就做轨迹筛选，WebArena Infinity 上开源里最強。

#CLIFT #WebArena #agents #GPT-5.5

arXiv cs.AI@Yifan Zhang et al. (6)Source

Papers BuildersSelected10:22

Growing Harness Turns Task Feedback into Reusable Agent Code

Moving recurring control out of context and into code means smaller models can run competent agents at a fraction of the inference cost, and the grown harness accumulates capability as reusable specialist code rather than prompts. Teams building agents on cost-sensitive or small-model deployments can apply this failure-guided repair loop directly, especially where Tool-Calling baselines collapse at the 4B scale. The ablations on trace-local edits, joint repair and gate rollback give a concrete recipe for anyone attempting similar harness-growing systems.

#Growing Harness #LLM #agents #inference cost

Event

arXiv cs.AI@Laizhen Li et al. (7)Source