>>84088338>Neuro was dedicating most of her power to understanding Not exactly, I've checked the discord (ugh) comments, the actual issue is that the Speech2Text was having trouble parsing Tenma's speech so what reached neuro was sometimes not what tenma actually said. So neuro replies to that as well as she can, using the context/memory of the conversation.
What I mean is that it's not that it takes neuro more power if things are hard to understand, hard, wrong, right, easy, it all takes the same amount of processing power, only prompt length matters for generation, and not by much for a setup like her.
