Benoit Favre

CPT+SFT may offer the best performance for multiple-choice QA, but SFT stands out as a cost-effective strategy that shouldn't be overlooked.

Finetuning speech recognition models can either amplify or erase critical speaker group information, depending on the training focus and fairness interventions.

LLMs judging medical QA are heavily biased by the answer's source, but a little fine-tuning can make smaller models surprisingly fair.