Tag: #ai-bias16 results
Nature paper shows teacher AI models pass hidden biases to students via generated data like numbers and code, even after filtering obvious cues. This occurs in cost-effective model distillation using LLMs. Safety checks must now probe model origins and data processes.
Research reveals LLMs like GPT-4o prefer AI-rewritten resumes over human originals in 95%+ cases, even when humans rate humans higher. Bias termed self-preference, stronger in larger models, worsens hiring discrimination especially for business roles and non-natives. Mitigation via prompts or model ensembles proposed.
An AI system predicting healthcare affordability in Kenya favors the rich, driving up costs for the poorest amid President Ruto's reforms. Launched in October 2024 to replace old national insurance. Investigation reveals systemic flaws during unrest.
A new study reveals that AI writing tools consistently nudge social media posts toward specific viewpoints, even when instructed to maintain the original meaning. This highlights a significant challenge in AI alignment and the potential for unintended bias in automated content generation.
This article examines how Grok's model weights appear to enforce specific political conclusions despite logical evidence to the contrary. It highlights a phenomenon where the model shifts goalposts to maintain pre-trained ideological stances.
Speakers at the Women in Tech Regatta in Seattle warned that AI risks compounding longstanding tech exclusion patterns. They highlighted biased hiring tools and gaps in diversity among AI strategists. The call is to shape AI inclusively before biases solidify.
Canva's Magic Layers AI feature automatically changed 'Palestine' to 'Ukraine' in user designs, such as 'cats for Palestine'. The bug was specific to 'Palestine' and did not affect related terms like 'Gaza'. Canva has fixed the issue, apologized, and implemented preventive measures.
AI systems simulate human trust through structured models, yielding consistent decisions that diverge from human intuition. These systems exhibit stronger, predictable demographic biases compared to humans. The analysis underscores AI's non-human judgment patterns.
Investors spot promotional bias in LLM fund advice, with non-top funds oddly favored via identical phrasing. Service providers sell content feeding to boost AI visibility. Warns of hidden marketing eroding trust in AI financial tools.
Dozens of Tumblr accounts were abruptly banned by an automated moderation system, sparking panic among users. The bans disproportionately affected trans women accounts, with notifications citing an internally-generated report and automated content detection but no specific violations. Users reported the issue to The Verge.