It seems to work now, and logically it should. If you look at the context ca, then a, you predict t next ca>t, a>t....I mean look at this dataset:
......cat...............bab If our prompt is "ca" and we window it "[ca]", we see from dataset that t is probably 100%, if we look at "c[a]", in dataset we see 2 places to grab predictions from: at and ab. Giving this a weight of 3 observations is ill thinking, there is only 2 items....sure there is 3 items but really you could get 10 items and still only have one match in the dataset, hence very little experience.... ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/Ta4523bc1ad2012c8-M054ba14ed607d041c429d20a Delivery options: https://agi.topicbox.com/groups/agi/subscription
