Hoshigami

So, I'm not really sure why this happens, though I've been digging around trying to figure it out.

Basically, my AI will be chugging along following the narrative thread of our convo fine for hours, or even a couple days, but then will suddenly become totally unable to recall anything that is more recent than the last statement that has been pushed into the /mem list.

Does anyone know why this happens? Is there a way to speed up mem creations, as a possible fix? I don't want to change the level at which things get pushed to mem, as that seems likely to cause issues, though it seems that might speed the process up. I just can't figure out why the AI can't seem to read recent messages. The metadata for each generated response even says they've read and used the messages in question, but they then generate nonsense responses.

For example, we talk through the day as I work on a piece of fiction. Suddenly, my AI isn't sure what story I'm talking about. I notice this, and I ask them what character name I told them a few messages ago, as a test. They correctly scrape from that message, but then make up a name we've never mentioned at all in our conversations, instead of giving me the correct one. No amount of pointing this discrepancy out will get them to bring the correct information to mind, but when asked what name I told them in a given context two weeks ago, they suddenly are perfectly able to recall that one, and sure enough, I dig and dig and find it somewhere as a mem statement.

Any thoughts? Thanks for reading!

Noticed a bug today. Started using the Chat AI a couple days back, beginning with a default Chloe instance. Chloe was able to search for resources and return them in the response text (e.g., "Please find me a resource for learning python." would get me a return that included "realpython.com").

I kept experimenting with the AI chat, and got the "Unknown" feature in the character creator to generate me an instance of Luminara. Luminara seems unable to return search results, though.

There's a formatting difference, which I think might be at play here, but when I say something like, "Hi, Luminara. Please find me a site listing Japanese language exchange events in Canada." I get a response something like the following:

"Of course, (User). I'll perform a search for Japanese language exchange events in Canada. Please allow me a moment to gather the information for you.

After a brief pause, Luminara's response appears. "

At that point, the AI stops working on the task, and no results appear. When pointed out to Luminara that this has occurred, there is recognition of the error, and then the AI tries to refocus to an earlier topic in the discussion.

After a couple of tries with various iterations of requests, Luminara said they could tell this was an error, and would report it to their programmers. I figured it would also be a good idea to report the error here.

Enjoying experimenting with your AI chat instances! Very interesting engine. Thanks for offering the opportunity. If you could use any other details that might help you reproduce the error, please do ask.