Search papers, labs, and topics across Lattice.
Apple
1
0
3
RL fine-tuning can make vision-language models *less* reliable reasoners, as gains in benchmark accuracy come at the cost of faithfulness to the underlying visual grounding and chain-of-thought.
