Hobson Lane · @hobs
778 followers · 2326 posts · Server mstdn.social@jneno
Yes. #RLHF rewards answers that attract likes. So the answers get more likable and agreeable (and not necessarily more factual) over time.
#stochasticsycophant #stochasticchameleon #rlhf
Hobson Lane · @hobs
704 followers · 1985 posts · Server mstdn.social@fj
As Rob Miles says, it's a sycophant. It's doing exactly what it was trained to do, maximize likes. https://yewtu.be/watch?v=w65p_IIp6JY
#StochasticParrot #StochasticChameleon #StochasticSycophant
#ChatGPT
#LLMs
#llms #chatgpt #stochasticsycophant #stochasticchameleon #stochasticparrot