The Download: gig workers training humanoids, and better AI benchmarks
Why it matters: Thousands of gig workers are directly shaping the capabilities of future humanoid robots, while new AI benchmarks are urgently needed to accurately assess AI's real-world impact.
- Gig workers in over 50 countries, including Nigeria and India, are recording themselves doing chores to train humanoid robots for companies like Micro1, earning locally competitive wages.
- Micro1 sells the collected data to robotics firms, making these videos the "hottest new way" to train humanoids, but sparking ethical debates around privacy and informed consent.
- Angela Aristidou and other experts argue that current AI benchmarks are "broken" because they evaluate AI in isolation, proposing a new "Human–AI, Context-Specific Evaluation" to assess AI within real-world human teams and workflows.
- OpenAI has secured Silicon Valley's largest-ever funding round, raising $122 billion ahead of its IPO, signaling massive investment in the AI sector.
- Infleqtion is competing for a $5 million prize with its quantum computer, aiming to solve healthcare problems that classical computers cannot, demonstrating the ongoing race in quantum computing applications.
Gig workers globally are training humanoid robots by recording daily chores, a method that is quickly becoming the primary way to gather data for robotics firms like Micro1, despite raising significant privacy and consent concerns. This innovative training approach highlights a critical need for new AI benchmarks that assess performance within complex human-AI environments, rather than isolated problems, to better understand AI's real-world capabilities and risks.
