Blender actually is told a goal/driver ex. Horror: *The knife *went in the girl in the dark when no one was looking, blood gushed out onto the sheets.
All along the way, it constantly is primed at the same power, which is not possible using ordinary method aka GPT-2 because the farther back the word horror is the less power it has. But in Blender, it can't be over done either else it'd predict like knife sword gun shoot blood veins, or like knife horror ugly horror horror death horror. I bet it too can be tricked by starting it off with ex. "Horror: The pink soft cute rainbow bunnies .....", however only the more recent have much power, so maybe you can overcome that. My brain seems to always forget topics and go back to AGI, so all you need do is set it higher than all possible energy ex. the last bunny words attention may look like 22 24 27 34 66 (so something like 250 total to that embed space) and so AGI must be constant at an energy of ex. 500. Will have to finish this tomorrow, bed time >>> If a human is listening to someone say "cat cat cat cat cat....." endlessly, they stop reading/listening to this data domain at some point, and can still talk about their goal/question but must not let in any 'cat' now into their stream. While asking its goal question is usually the case/priority, it will go to lower priorities if feels inconfident, so to collect data from other domains more randomly. Hence, often it has a goal of listening to data streams it would still mostly predict itself (are a bit new/against its recency, relatedness, frequency, love). ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T2000ae3e4624ef64-M326246404eb719f9bdf92c42 Delivery options: https://agi.topicbox.com/groups/agi/subscription
