After creating new conversations the AI is still referencing content from previous conversations. New conversations should be a complete clean slate.
Example: I had one chat about downloading a youtube video. The AI was able to install a python plugin and fetch the video.
In a completely different conversation I asked it to download a URL. It installed and used the same YouTube download plugin, unprompted and without any context to YouTube.
This is not the only example and the cases are not specifically related to YouTube or URL downloads. This is a general problem that needs to be looked into.
What you are trying to do with julius is not ok with me. Julius is for computational purpose.There are many guides in this forum and also in the Julius web how to use it ( use cases) and how your data should be formatted before you use it.
In the second convo, did you ask it to download the URL though? because Julius uses python to do things and it would naturally load the same library to accomplish this, and the URL would contain the word youtube in it, enough for Julius to deduce what to do next.
Hey @joshua, do you mind sharing the chat links of the two conversations you’re referring to? As of April 2024, Julius doesn’t have memory that transfers between conversations so it’s either a random coincidence or the AI concluded that the youtube module is the right way to download data from the link you shared. Either way, happy to look into this further.
If you prefer to share the chats privately, feel free to email us at team @ julius.ai and we can take a look.
Have had the same issue of contaminated conversation context when editing previous messages.
Often times I’ll just start from a previous message in the thread when Julius goes off track, and have observed it using variables or other context downstream that it shouldn’t know at that conversation checkpoint.
Maybe the contamination comes from the python environment, so unintended context is injected during the “Examining resources” process?