although having said that, even simpler probes require some operationalisation of the target state (e.g the model is lying) which is normally behavioural rather than ‘bottom up’ (lying requires believing things, which is an intentional state again.)
although having said that, even simpler probes require some operationalisation of the target state (e.g the model is lying) which is normally behavioural rather than ‘bottom up’ (lying requires believing things, which is an intentional state again.)