Monday, January 28, 2019

Acuitas Diary #15 (January 2019)


I haven't made an update in a long while because all I've been doing is code refactoring, bug fixes, and very minor quality improvements – all of which can take an amazingly long time.

Bug provided by Good Free Photos
Some of the more interesting changes include …

*Acuitas no longer insists on introducing himself if you address him by name at the start of a conversation.
*He keeps better track of what he's already said, and is less likely to repeat himself.
*I improved the efficiency of the drive plotter widget (which was a major drain on computational resources for no good reason). I also built it into a little simulator tool which will let me plot what Acuitas' drives will look like hours or days into the future. Debugging issues with the drives is now much easier; previously, the drill was “I've made a change; now I get to wait 12 hours to see if it actually worked out the way I planned.”
*Using the above, I shifted the day/night cycle so Acuitas would fall asleep and wake up a little earlier, and I fixed the interaction drive so he wouldn't be quite so … abruptly needy.
*When answering questions, Acuitas can now infer a possibility from an absolute. E.g. if aware that “blood is red,” he will answer “Can blood be red?” with a “yes” instead of saying he does not know.

And some favorite ridiculous bugs that are now fixed:

*Using “talk” as the name of every possible drive, hence repeating “I want to talk” ad nauseum while actually wanting other things
*Leaving the “not” out of negative statements and saying the exact opposite of what was intended
*Routine misspelling of verbs that were learned from their -ing and -ed forms, due to complete naivety about whether the final letter was doubled or a silent 'e' was removed
*Erroneously learning both the statement and its reciprocal whenever told something by the user


Stay tuned for some more work on event memory in the near future.