The bidding function is probably adding a lot of latency for little benefit.
One case where it's helpful is to ensure that when spoken to, the specific Plantoid is more likely to respond. But even this requires some prompt tuning.
Should we re-factor? It may be helpful to have simpler selector functions, that determine turns either randomly or in a sequential order. Some of this is probably implemented, please advise G...
We have de facto replaced LLM bidding with random speaker selection, excluding the last speaker. Bidding hooks by LLM still exist if we want to bring them back.
The bidding function is probably adding a lot of latency for little benefit.
One case where it's helpful is to ensure that when spoken to, the specific Plantoid is more likely to respond. But even this requires some prompt tuning.
Should we re-factor? It may be helpful to have simpler selector functions, that determine turns either randomly or in a sequential order. Some of this is probably implemented, please advise G...