Ben Gilburt
- talk
Looking for a Handsome Carpenter! Measuring and Mitigating Bias in LLMs
Learn from Ben" Prompt engineering is ineffective at reducing bias. It slso doesn't create high quality outputs. The best thing you can do is fine tune based on low bias data. It tends to be that if it's low bias, it is higher quality data"
The growing capability and availability of generative language models has enabled a wide range of new downstream tasks. Historical research has identified, quantified and mitigated biases present in language models but is rarely tailored to downstream tasks where wider impact on individuals and society can be felt.
This presentation shows a real-world impact of bias in language models in a situation that many us engage with throughout our lives, job ads. It quantifies the levels of bias present in language models, and appraises the most effective methods for mitigating bias in the workplace, so that the outputs of LLMs can be more inclusive.
