ResearchResearch paper

PrefDiag-Bench diagnoses how and when personalized GUI agents fail

Submitted October 22, PrefDiag-Bench uses 318 end-to-end tasks and 1,664 diagnostic probes to trace how preferences alter execution and when they should activate in GUI state; all 11 systems trail their own non-personalized baselines.

Event 2026-10-22Source published 2026-10-22Indexed 2026-10-23Verified 2026-10-23

Why it matters

Generative and agent interfaces must do more than remember a preference: they must select the right historical evidence and activate it at the right interface state. Otherwise personalization becomes faulty automation at ranking, process or confirmation boundaries.

A closer look

The benchmark groups preferences into Outcome, Constraint, Ranking, Process and Interaction, comparing explicit profiles with behavioral-memory logs. The strongest system, Seed-2.0-Pro, scores 81.13% on basic tasks but averages 62.64% with profiles and 65.66% with logs; its Process score under profiles is 47.17%. On an 85-task subset, labeling logs as supporting evidence or noise raises Seed-2.0-Pro from 76.47% to 83.53%, while removing noise drops it back to 81.18%.

Who this is for

Teams building personalized agents, long-term memory, preference triggers, confirmation flows and GUI-automation evaluation

Original sources

Revisions & corrections (1)

2026-10-23 — Paper, full text and official repository verify date, benchmark size, results and construction method; caveats preserve synthetic-context and intervention limits.

Related events

Research2026-10-29

A second, independent VeriCUA turns GUI-task acceptance into active evidence gathering with abstention

A Google and University of Hong Kong team submitted a separate project also named VeriCUA on October 29. It inspects an environment or recording before constructing a task rubric, then gathers missing evidence through GUI, accessibility-tree, file and database reads, marking requirements met, unmet or unknown. In online experiments it reduces false positives against baselines on AndroidWorld, OSWorld and ScienceBoard.

Google and University of Hong Kong VeriCUA arXiv paperResearch paper

Research2026-10-27

GUI‑JEPA replaces token-by-token action decoding with candidate-conditioned latent scoring

Submitted October 27, GUI‑JEPA targets single-step mobile GUI action selection. It transforms JEPA representations pretrained on unlabeled interface screenshots for each action and region candidate, then scores accessibility-tree candidates in one batch instead of asking a 7B vision-language model to decode an action token by token.

GUI-JEPA authors / arXiv and official repositoryResearch paper

Research2026-10-27

TAM turns verified task advancement into online state and reusable operations for GUI agents

Submitted October 27, Task Advancement Memory (TAM) uses what has been reliably accomplished at two timescales: it maintains verified progress, remaining goals and visual evidence during a task, then extracts reusable operations from locally successful segments after the task, including segments inside an overall failed workflow.

arXiv paper and full textResearch paper