Fixing Failures in Browser-Use Models: Why More Data Isn't Enough
jiggle123
17 points
1 comment
June 26, 2026
Related Discussions
Found 5 related stories in 974.1ms across 14,015 title embeddings via pgvector HNSW
- GUI-Perturbed: Breaking Browser-Use Models Using Domain Randomization jiggle123 · 12 pts · June 24, 2026 · 64% similar
- Better Models: Worse Tools leemoore · 138 pts · July 04, 2026 · 61% similar
- Smalltalk's Browser: Unbeatable, yet Not Enough mpweiher · 139 pts · March 05, 2026 · 57% similar
- Show HN: Pulpie – Models for Cleaning the Web snyy · 86 pts · July 06, 2026 · 53% similar
- What if the browser built the UI for you? jonnonz · 22 pts · April 05, 2026 · 52% similar
Discussion Highlights (1 comments)
dippogriff
Great work showing on how brittle these GUI benchmarks can be! Love the visuals. I wonder if SFT is the problem here as opposed to the coordinate discretization; what happens with continuous action space?