Search papers, labs, and topics across Lattice.
Predictive Labs Ltd, University of Oxford
2
0
4
Randomized splits can inflate model performance by up to 6.5 times, revealing the critical need for temporal leakage audits in financial NLP.
Salience Bias in LLMs reveals that models often ignore commonsense reasoning in favor of misleading explicit cues, with lightweight prompting showing promise in addressing this issue.
