Learn on the fly

[PiperMp3]

Hello,

I've currently only gotten so far to have her installed, running as CLI and tested Web App a little. I've yet to see the training of the neural network during a new training.json occur. What I've read is when new intents or new datasets are being added Olivia needs a quick nap (restart) . How could we get her to never need a restart and can detect file changes / new datasets.. new intents, during runtime and she will automatically decide to start training. She could still process anything asked during this training that she already knows, but she will announce to the user that she has detected new data and is currently learning. When she's done she will also state the completion and at this point the new information she's learned is available to be requested/answered with.

Would such a feature be possible?

The other method that would also be interactive is that the user could ask Olivia, "Why don't you do a little homework?"(or something) Whereas Olivia remains active, running, but she rewrites her training.json using the files she has at the time.

Mighty impressed by your work so far!

Side question, as I didn't really see it mentioned, or perhaps I missed it.. 1. Does Olivia have a limit to how many intents / datasets she can handle? If I copied Wikipedia into intents and sets, giving her 800gb of data, how would she handle it?
2. During runtime, is Olivia's "brain" running in RAM or is she pulling information to make answers from file system on demand?

Many thanks, and keep at it. Good work !!

[hugolgst]

Hello,

It is definitely possible to have a such feature, I only did the current one to save time.
We could inject some features and then just re-train the model a bit to include it.

Technically she could handle it but I think that it will take a long time, during runtime there is a save of the neural network loaded into the RAM. There is also a cache if you ask multiple times the same question.

Thanks a lot for that, I appreciate it.

[PiperMp3]

Good morning,

Many thanks for coming back to me - I saw mentioned that a Discord server was started?
I have some further questions and I don't want to spam the GitHub issues page :)

I understand Olivia does not store any data, that may come from the Swiss culture.
Would be cool to have an optional setting where Olivia could track an unique identifier to see whether its "me" or someone else, and store data accordingly.

An example; I talk to Olivia as ME, I share secrets (perhaps) and then my partner connects as her (different name, perhaps different device ID that Olivia now sees) she will say "Hi [NAME]!" and Olivia now knows not to talk about me, but she now uses all data she knows about the other person. Should I ask Olivia to share secrets that my partner has told her, Olivia would refuse etc.

How could one implement such option? Ever had a similar thought?
Currently I got Olivia via CLI, but depending on how she's evolving and how I can teach her more I would most likely look to eventually talk to her, to extend her speech - How could wouldn't it be that she recognizes different voices? I have one voice, my partner has another and Olivia answers us accordingly.

[hugolgst]

I think to do such a feature Olivia would need to share the ID of the other user and make the client replace it in its local storage.
Of course it can be done but I'm not sure that it is in the spirit of Olivia, you could however implement it in a fork :)

[A-Yamout]

I think that is great that we value the privacy of others and keep expanding Olivia. @hugolgst you have done a great job with this and I would love to help if there is any way that I could help with the new 3.00 release just ask.

BTW I am great with logos graphics and art more then go (I am horrible at go) so want a new logo or a new UI for the web app I can help you with that, and thanks for all your hard work.

[hugolgst]

Thanks a lot for that! However, it is hard for me to keep maintaining Olivia thus I am pretty busy.
I am very open to new ideas so you can create issues or work on some PR if you want so!
Huge thanks! :)

[A-Yamout]

No Problem. Thanks for making this open source and open to the world.