3 ms·
> Isn't token completion incapable of representing AGI? Given the absence of a validated model (or even usable operational definition) of general intelligence,
by dragonwriter 3y ago
> Isn't token completion incapable of representing AGI?
Given the absence of a validated model (or even usable operational definition) of general intelligence, who knows? AGI might as well be an empty marketing buzzword, it isn't something about which falsifiable fact claims can be made.
> AGIs need the ability to perform internal thought and deliberation
Systems built around LLMs can do something like this, via reflection, a technique in constructing agent-based simulations using LLMs as the engine.