VLA Trained on Bridge Data Fails to Recognize a Hammer
svlevine · x · 2026-07-04
Sergey Levine illustrates a problem with an example: a VLA model trained on the Bridge dataset doesn't know what a "hammer" is. When asked to put a hammer on a plate to prepare an "iron-rich meal," it grabbed a spoon instead. While a reasonable guess, this exposes the model's flaws in semantic object understanding.
More from Research
- AI folks, stop saying "biology" when you mean drug development — KevinKaichuang · 2026-10-26
- ostris pre-training "Primal VAE": 1-channel latents match modern 8x VAEs via patching — ostrisai · 2026-10-26
- Shanghai AI Lab-led team releases 100-hour egocentric wearable dataset for humanoid training — ChuckDBrooks · 2026-10-26
- ConvFill: A 0.6B Talker Buys Time While a Large Reasoner Whispers the Real Answer — andimarafioti · 2026-10-26
- Flow Reasoning Models turn flow matching into iterative reasoners, hitting 99.5% on Sudoku-Extreme — alec_helbling · 2026-10-26
- Surprising find: Alzheimer's-linked amyloid fibrils deliver RNA to sperm for embryo translation — EricTopol · 2026-10-26