Large Language Models (LLMs) frequently reproduce the gender- and sexual-identity prejudices embedded in their training corpora, leading to outputs that marginalize LGBTQIA+ users. Hence, reducing such biases is of great importance. To achieve this, we evaluate two parameter-efficient fine-tuning (PEFT) techniques - Low-Rank Adaptation (LoRA) and soft-prompt tuning - as lightweight alternatives to full-model fine-tuning for mitigating such biases. Using the WinoQueer benchmark, we quantify bias in three open-source LLMs and observe baseline bias scores reaching up to 98 (out of 100) across a range of queer identities defined by gender and/or sexual orientation, where 50 would indicate neutrality. Fine-tuning with LoRA (< 0.1% additional parameters) on a curated QueerNews corpus reduces those scores by up to 50 points and raises neutrality from virtually 0% to as much as 36%. Soft-prompt tuning (10 virtual tokens) delivers only marginal improvements. These findings show that LoRA can deliver meaningful fairness gains with minimal computation. We advocate broader adoption of community-informed PEFT, the creation of larger queer-authored corpora, and richer evaluation suites beyond WinoQueer, coupled with ongoing audits to keep LLMs inclusive.

Israeli researcher Lena Neufeld examines a 3D model of a malignant tumor, as part of a brain cancer research that uses patients' cells to make 3D printed models of tumours, at Tel Aviv University, Israel August 17, 2021. Picture taken August 17, 2021. Israel and Germany have launched a joint forum to work on advancing the use of artificial intelligence and machine learning in healthcare, with the two nations keen to learn together from the lessons of the coronavirus pandemic. As part of a three-year project, the German Israeli Health Forum for Artificial Intelligence will bring together stakeholders from the health ecosystem, startups and experts of both countries to discuss developments, regulations and applications of AI solutions in the health sector. The initiative is funded by the German Federal Ministry of Health and was established together with the European Leadership Network (Elnet).

In May 2017, the LGBT organization GLAAD posted a video on YouTube of actress Debra Messing receiving an award. In Messing's acceptance speech, she praised many Americans for supporting one another and reminisced about the show she starred in, NBC's CMCSA, 0.36% "Will and Grace," and its influence on telling the stories of members of the gay community. She also called on members of the Trump administration to "do right" by the LGBT community, by removing Steve Bannon (who has since left) from his post as President Donald Trump's chief strategist. She did not specify what her criticism of Bannon was. She also said in her speech that Ivanka Trump should work for "women's issues."

The accelerating pace of digitization is bringing real, tangible benefits to our society and economy, which we cover daily in the pages on this site. But increased reliance on machine learning algorithms brings its own unique set of risks that threaten to unwind progress and turn people against one another. Three speakers at last week's Strata Data Conference in New York put in all in perspective. The first Strata speaker with a dire warning about the potential for algorithms to go haywire was Cathy O'Neil, who holds a PhD in mathematics from Harvard University and is the founder of the website mathbabe.org. After working as a quant in the finance industry for several years, O'Neil became disenfranchised with the field and how it used data science to rack up profits at customers' expense.