Sign In
Hiroshi Goto
Why is Google Gemini more bias-affected rather than ChatGPT ?
Gemini and ChatGPT both utilize transformer-based LLMs. However, the way of reinforcement learning is fundamentally different. ChatGPT employs reinforcement learning from human feedback (RLHF) as we've known well. However, Gemini employs reinforcement learning from non-human agent feedback. This agent determines feedback of reward or penalty based on deep neural network or statistical curve fitting results. That's why Gemini suffers bias in the same level of conventional deep learning. RLHF is applied to Gemini only when a customized system is required because RLHF is a double-edged sword concerning cost and performance balance. You may understand the background of this article.
Google Gemini: Former employee, tech leaders suggest what went wrong with the AI chatbot