Jiaming Wei

Jiaming Wei, Zekun Wu, Adriano Koshiyama, María Pérez-Ortiz

Accepted at EMNLP 2026 Workshop REALM

Submitted to NeurIPS 2026 Workshop VLM4RWD · under review · arXiv:2608.06171

• Controlled comparison of DOM / SoM / vision and phantom representations, preregistered before the main runs.
• Core result: a routing value–learnability gap — value does not imply learnability.
• Tracks task-level failures and representation differences, with activation patching / linear probes for diagnosis.
• Large reproducible research system with heterogeneous-compute orchestration, automatic recovery and 1K+ research tests.

MSc Artificial Intelligence for Sustainable Development · Department of Computer Science

Programme completion: 21 Sep 2026 · dissertation result pending

Primary supervisor: María Pérez-Ortiz · Second supervisor: Zekun Wu

MSc dissertation: When Is Expensive Perception Worth Paying For? Measuring the Ceiling, the Predictability and the Economics of Representation Routing in Web Agents

BEng Automation · GPA 3.82 · weighted average 87.95

• Owned an end-to-end red-team/evaluation measurement line and separated target behavior, provider-side filtering and grader outcomes instead of collapsing them into one ambiguous binary result.
• Audited judge/grader FP/FN and failure modes against independent labels, then integrated the measurement logic into the working platform.
• Later built an NYC Local Law 144 audit pipeline: deterministic Python decision logic, evidence/logging and reviewable reporting, validated on real data and demonstrated live at the end of the internship.

Puts the target, execution harness, provider behavior, judge and independent gold labels on the same measurement chain. The system asks not only whether the target failed, but whether the evaluator itself is trustworthy.

• 166 plugins mapped to local generation mechanisms; 103 enabled after end-to-end verification.
• Target-conditioned attack generation; a zero-egress path can keep generation, execution and judging local.
• Found an upstream refusal shortcut that missed 8 of 16 breaches on one probe.

Controls protocol, answer-format SFT, explicit reasoning and RLVR separately to attribute score changes to capability, protocol alignment or reward exploitation.

• Switching back to the correct native chat protocol alone recovered roughly +20–30 points for some larger base models.
• 4B answer-only SFT showed no significant net gain; paired McNemar turned an apparent improvement into a testable claim.
• GRPO/RLVR exposed clear reward exploitation while task accuracy fell.

English / tests: TOEFL iBT 107 (R28/L28/S23/W28) · MyBest 109 · 21 Sep 2024 · GRE 321 excluding writing