Tag: #data-annotation19 results
Meta ended its contract with Kenyan AI firm Sama after workers reported exposure to disturbing graphic footage from its AI smart glasses. The lawsuit highlights ethical concerns in AI training data collection. This incident questions the safety of real-world AI data sourcing.
Meta faces backlash after over 1,000 Kenya-based workers were laid off by a subcontractor. Workers claim they saw smart glasses users having sex in reviewed footage. Meta and the subcontractor dispute the redundancy reasons.
Sama will lay off over 1,100 workers in Kenya following the end of its Meta contract. The cuts represent a significant blow to Kenya's fast-growing but fragile AI outsourcing sector.
The article delves into the peculiar human labor behind training AI for human-like conversations. Workers vent emotions, confess secrets, and role-play with strangers to create training data. This process helps machines mimic natural speech patterns.
SynthyfAI turns image/text labeling into a game: guess popular answers from prior players per round. Level up customizable 'AI' synth characters as you annotate. Early prototype shared for community feedback.
Freelancer Katya joins Mercor after a sketchy LinkedIn offer, creating prompts, responses, and checklists to train undisclosed AI models. She enjoys good pay and real-job feel initially. Project halts after two days, leaving her jobless amid financial desperation.
Field study reveals China's 'inland-sourcing' model for AI data annotation in rural relocation communities, prioritizing data security over global outsourcing. Bases employ mostly local women in factory-like settings for stable, high-accuracy labeling of images, speech, and text. Local governments partner with tech firms for poverty alleviation and employment stability.
Workers manually label everyday objects in photos and videos to train AI systems. This data annotation process enables AI to accurately identify common items. It's a key step in developing computer vision capabilities.
Literature review critiques 'ground truth' in ML data annotation as a positivistic fallacy ignoring human subjectivity. Analyzes 346 papers from top venues revealing biases like anchoring and geographic hegemony. Proposes roadmap for pluralistic infrastructures embracing disagreement.