• vaderaj
    link
    fedilink
    English
    arrow-up
    1
    ·
    1 hour ago

    What exactly makes you say, LLMs are taking decisions? From what I understand it plays a probability game of predicting next set of words.

    From those words chosen there is ReAct(Reason and Act) framework which selects the best possible next executable task from the choices provided to it.

    For all I know, decisions are thought out actions using a logical flow and not predicting the next best token based on the previous token. And for every decision we are held accountable brute forcing our way to make decision is not viable IMO. Would like to hear other perspectives or would like to know if I missed something.

    https://www.ibm.com/think/topics/react-agent