“It actually does caching of prompts so that if you do the prompt before, it doesn't test it again, saves you money”

“It writes a prompt… submits that prompt to an LLM… the agent program just directly executes whatever it says… and then it loops”

“We can go into the KV cache of the model and steer these positive or negative directions ... when you steer the zone with negative valence, the model is significantly more likely to avoid that zone”

“We just introduced Dreamforce on-device transcription… it will actually populate your visit based on that conversation and say, they wanted samples… Do you want to send it?”

“If you come back after one hour and one minute, that cache is gone… cache writes are expensive… leaving sessions idle for more than an hour is expensive.”