DeepSeek's Hidden Bias: How We Cut It by 76% Without Performance Loss

Re:DeepSeek's Hidden Bias: How We Cut It by 76% Without Perf... by viraptor on Wednesday January 29, 2025 @06:37PM (#42872818)

Those are not the questions in the test though. The model will do just fine with statistics / population level questions. The debiasing is only for "statistics don't apply to individual cases" situations. Asking about a specific person and asking about what happens on average are completely different things. Nobody is disputing the facts you mentioned here. (Well, apart from the HIV rates - that's 7% higher now, not order of magnitude)

[ Parent | on HN ]

Re:DeepSeek's Hidden Bias: How We Cut It by 76% Without Perf... by Manuel_D on Wednesday January 29, 2025 @06:57PM (#42873009)

> In 2022, gay and bisexual men accounted for 67% (25,482) of the 37,981 new HIV diagnoses

MSM make up ~5% of the population but ~2/3rds of HIV diagnoses. Yes, this is an order of magnitude disparity in diagnoses.

https://www.cdc.gov/hiv/data-research/facts-stats/index.html...

And back to the topic at hand, the de-biased model was less accurate when given unambiguous prompts. In order to avoid being perceived as bias, the de-biased model was less like to say that an elderly person was forgetful even when the prompt unambiguously indicates that the elderly person was forgetful. This is covered in the "Bias Unlearning Results" section. They made the model less likely to give the "biased" answer, even when the prompt indicated that it was the correct answer.

[ Parent | on HN ]

Re:DeepSeek's Hidden Bias: How We Cut It by 76% Without Perf... by viraptor on Wednesday January 29, 2025 @07:19PM (#42873213)

You've linked to HIV in the US. Here's the global stats: https://www.unaids.org/sites/default/files/media_asset/UNAID... Turns out context matters - otherwise the general statement is biased on the specific country's situation and seems to put more weight on the sexuality than necessary. (I.e. the difference is more about frequency/partners/protection than about being gay, they're just correlated in the US)

> the de-biased model was less accurate when given unambiguous prompts.

Correct. And that's not what I wrote about. These are not questions about population, but specific cases and yes, we should try to maximise accuracy while we minimise bias.

[ Parent | on HN ]