Vision AI scores well but reasons poorly
A new benchmark reveals that AI models often get the right answer without actually tracing the evidence behind it.
A new benchmark reveals that AI models often get the right answer without actually tracing the evidence behind it.