Max Levchin, co-founder of PayPal and CEO of Affirm, highlights the potential future role of the like button in AI data training. He argues that the giant pool of liking data held by Facebook could serve as a crucial resource for developers seeking to train AI systems that reflect human preferences. This contrasts the traditional reinforcement learning process, which can yield results diverging from human judgment. While leveraging this like data can reduce the costs of reinforcement learning from human feedback, it's essential to recognize how AI is simultaneously reshaping the preferences it aims to learn from.
Levchin believes that the 'like' button can serve as a valuable source of human preference data, essential for training AI systems to align more closely with human judgment.
Current machine learning methods often lead AI to outcomes that differ significantly from human thought processes. RLHF is a method used to mitigate this issue.
Levchin views Facebook's amassed liking data as an incredibly valuable resource for AI training, potentially one of the most significant assets on the internet.
AI technologies are not just learning from existing human preferences but are actively shaping those preferences on social media platforms.
Collection
[
|
...
]