Replying to Avatar atyh

anyone else notice how easy it is to confuse chatgpt into incorrect answers?

it seems to contextualizes answers based on the conversation, and builds on each answer as a truth, eventually including incorrect things based on 'temporary absolutes' it created in the conversation string.

but if you ask the same question out of context in a different conversation, it will give the right answer.

its interesting.
Avatar
Ideia da Coerência - Conceito 1y ago

Yes, AI mimics natural language, seeking coherence rather than logical consistency; and, more than this, when it makes a mistake, it tends to equivocate and try to convince us of the opposite of what we know to be wrong, at least at first; AI is a kind of sophisticated sophist, almost human, all too human.

Reply to this note

Please Login to reply.

Discussion

Avatar
atyh 1y ago

but even this argument attributes intent. which the software lacks.

it likely has to do with the tokenization process.

Avatar
Ideia da Coerência - Conceito 1y ago

ok, intentionality is a complicated concept; let's put it this way, AI does not 'seek,' but 'tends to' produce dynamic, relational, and context-dependent meaning, that is, coherence rather than mere logical consistency.