Consider the following: we don’t train LLM to optimize for a decision made on local context. That would cover a minimum viable manager-bot. We train on (admitt

Consider the following: we don’t train LLM to optimize for a decision made on local context. That would cover a minimum viable manager-bot. We train on (admittedly flawed) context and offer decisions. Much closer to an executive advice bot. If given two checkpoints: your competitions manager bot, or your competitions executive advice bot, the latter is much more valuable.