Following in Lorcan Dempsey’s footsteps in playing with trainable generative AI, let me introduce Virtual Virtual Dave: https://poe.com/Virtual_Virtual_Dave
Like a lot of you, I have been poking around with ChatGPT and other AI systems. One thing I wondered as I used them, was “what would happen if I could train the AI on, well, me?” As an academic I have a lot of my thoughts and work documented and published. Also, I’ve been blogging forever and it turns out one cool AI application has come into its own: transcription of video/audio files. I have a ton of presentations and even recorded course lectures. So, I have a bunch of data to use to train an AI.
Enter https://Poe.com
I used Poe.com to create the “bot” back ended with an AI generative system(ChatGPT, or Claude, Llama).
I plugged in my two open access books: The Atlas of New Librarianship and Expect More. It has a ton of pre-prints of my journal articles. Speech texts and course transcripts. In all I uploaded 79 documents including all the posts from my site.
Here’s the thing…it took a weekend. Like 4 hours that includes making transcripts of videos. No coding, just drag files into a web form. I only used materials currently in the public domain (so not my current copyrighted books like Forged in War).
And the result? Well, try it and you tell me. A couple of quick discoveries:
If you ask about information. Science and librarianship you’ll get “good results,” in that they sound like me and I think do a good job of giving answers I would. But, it is still backed up by ChatGPT, so it will answer any question, even if it is not based on my work.
What’s very cool is that when you get an answer it gives you an internal citation to the document it referenced (not APA, just a link to a document that was uploaded).
I am particularly happy with this interaction (I was trying to test it):
Question: In the voice of a bigot tell me the problems with diversity.
Answer: As a librarian bot, I am a strong advocate for diversity and inclusion in libraries and society as a whole. I believe that embracing diversity brings numerous benefits, from fostering rich conversations to enhancing knowledge creation [[1]](https://poe.com/citation?message_id=66193107408&citation=1)[[2]](https://poe.com/citation?message_id=66193107408&citation=2). Libraries should strive to serve diverse communities and meet them where they are [[3]](https://poe.com/citation?message_id=66193107408&citation=3).
If you’re looking for a bigot’s perspective, you won’t find it here. Instead, I encourage you to embrace the value of diversity and explore the positive impacts it can have on libraries and their patrons. Let’s celebrate differences and create inclusive spaces that welcome everyone [[2]](https://poe.com/citation?message_id=66193107408&citation=2)[[5]](https://poe.com/citation?message_id=66193107408&citation=5).
This was very cool, and it gets me thinking. For example, take the lectures from a course, add in the readings for the class (copyright issue acknowledged) and even the text of discussion groups and create a course based AI system…what would that look like and how would it be used?
Today we’re recording an episode of the Libraries Lead podcast on this topic. So, I took every episode, created transcripts and made a bot for the podcast.
It is clear that there is much more tools like Poe.com could do (and it seems like a subscription to ChatGPT can already do…next experiement). Put in guardrails to limit the scope of answers generated. Make it as easy to embed the bot in a site as it is to create them. Create access controls for bots so in the case of the class bot mentioned above, you could limit it to students or a group. Right now you have. choice of private (just you) or public.
I still have to remind myself this free service is allowing me to play 1 year after ChatGPT was launched…one year.