User Modeling – Spotify Research : Spotify Research The user modeling research at Spotify entails translating users’ in-app activities into human traits, interaction models, emotional understanding modules, and situational contexts, thereby uncovering our users’ individuality. To do this, we use a multidisciplinary scientific approach at the intersection of music psychology, behavioral analysis, and machine learning. This enables us to build datasets and user interaction models to create engaging and personalized experiences for our users, in a data-driven fashion based on users’ interactions and feedback signals.
The urgent need for accountability in predictive AI Unfortunately, companies have exploited public confusion around the disparate types of technologies that fall under the umbrella term “AI”. While some types of AI are rapidly advancing, predictive AI is not. This fact is not widely appreciated, whether by those who buy these tools or those subject to its decisions.
Statement on AI Risk | CAIS Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war.
AI Enforcement: Examining the Impact of AI on Judicial Fairness and Public Safety by Yi-Jen (Ian) Ho, Wael Jabr, Yifan Zhang :: SSRN More importantly, we find that the AI’s recommendations impact judicial fairness in two opposite directions. On the one hand, while judges are typically more lenient with female offenders than males, AI helps alleviate such gender-based disparity. On the other hand, as judges maintain fair sentencing without the presence of the recommendations, we find evidence of a racial bias favoring White offenders over Black ones with the recommendations.
Measuring Implicit Bias in Explicitly Unbiased Large Language Models Using these measures, we found pervasive stereotype biases mirroring those in society in 8 value-aligned models across 4 social categories (race, gender, religion, health) in 21 stereotypes (such as race and criminality, race and weapons, gender and science, age and
negativity). Our prompt-based LLM Implicit Bias measure correlates with existing language model embedding-based bias methods, but better predicts downstream behaviors measured by LLM Decision Bias. These new prompt-based measures draw from psychology’s long history of research into measuring stereotype biases based on purely observable behavior; they expose nuanced biases in proprietary value-aligned LLMs that appear unbiased according to standard benchmarks.
Few Americans Would Encourage a Young Person to Become a Teacher | NORC at the University of Chicago Fewer than one in five Americans (18 percent) would encourage a young person to become a K-12 teacher, citing low pay, lack of sufficient school resources, and a stressful work environment as major barriers, according to NORC at the University of Chicago. These findings come amid a national teacher shortage and are from the latest NORC Spotlight on Education survey powered by AmeriSpeak®.
30 Years in Edtech-D’Arcy Norman dot net One thing I’ve noticed is that the sense of creativity, curiosity, exploration, and playfulness that defined the multimedia and early Web eras feels to have been subdued over the last decade or so. Part of that is due to the maturity of software and tools. Part of it is the neoliberal erosion of institutions. Part of it is a product of a shift from a creator mindset to one of a consumer. Edtech is now commoditized, productized, sold to us. It promises efficiency and scale – so that we can better respond to things like restricted operational budgets – to do more with less.