Richard Mathenge was a team lead among contractors based in Nairobi, Kenya, where he trained OpenAI's GPT models. His role was with Sama, an AI training firm that collaborated on this initiative. In a recent episode of the Big Technology Podcast, Mathenge recounts his experiences and challenges. During the training phase, he often encountered deeply disturbing content, received inadequate support, and his colleagues sometimes earned as little as $1 per hour.
Mathenge led a group in Nairobi, Kenya, where they trained OpenAI's GPT models as contractors. His position was as a team leader for Sama, an AI startup involved in the project. In an episode of the Big Technology Podcast, he shares his story. Throughout the training, he was repeatedly exposed to sexually explicit material, lacked adequate counseling, and his teammates sometimes received wages as low as $1 an hour. Tune in to discover more about the intricacies of training these models and the human aspect behind Reinforcement Learning with Human Feedback.
2023
While at work, Mathenge and his team frequently tagged and assessed explicit content for the model. The origins of this material were often unclear, and they had to categorize it as erotic, child sexual abuse content, illegal, non-sexual, or other types. Much of what they had to read was shocking. Mathenge noted that some texts depicted horrifying scenes, such as child rape or bestiality in front of children. He opted not to disclose certain details due to their highly offensive nature.
The kind of labor Mathenge engaged in is critical for the performance and allure of AI creations like OpenAI's and Google's Bard. However, the human sacrifice behind this effort has largely gone unnoticed. This process, known as Reinforcement Learning from Human Feedback (RLHF), enhances bot intelligence by teaching them to improve based on user interactions. Prominent figures in AI, including OpenAI's Sam Altman, have praised the effectiveness of this technology, but they rarely acknowledge the human toll required to align AI systems with ethical standards. Mathenge and his colleagues found themselves directly facing the realities of this situation.bots likeChatGPT OpenAI has restored Sam Altman to his position as CEO after his sudden dismissal by the board.artificial intelligence The newly formed board, which now includes economist Larry Summers, Quora’s co-founder Adam D’Angelo, and a non-voting observer from Microsoft, will have Bret Taylor as its Chair and Microsoft’s leading investor. Altman expressed gratitude for Microsoft’s support and optimism.
Latest updates from Richard Mathenge on social media Content moderators endure graphic and disturbing material as part of their daily responsibilities to maintain safety in our online environments. Let's recognize their steadfastness and dedication to protecting digital standards.
Every online experience is supported by the diligent efforts of content moderators working to protect our virtual spaces.
Moderation guidelines vary significantly from one platform to another, making it essential for users to understand them. Let’s take the time to educate ourselves about the regulations governing the online landscape, so we can foster a more positive digital culture. #ModeratorsMatterpic.twitter.com/tAjPSfyuxt
