AI Chatbot Training: Your Own Thoughts
The image above is a snapshot of our website chatbot pop-up. The chatbots name is “Asyra” which sounds similar to “Asycd”.
Abstract
In today’s digital landscape, conversing with chatbots often feels impersonal and devoid of life. But what if we could change that? Imagine engaging with AI that not only responds but also resonates with depth and purpose, mirroring the intricacies of human thought.
At Asycd, we’re pioneering a new approach to chatbot interaction by infusing our AI with the essence of our collective thoughts and ideas. Every day, we embark on a journey of introspection, documenting our musings and plans for the future of Asycd. Not only Asycd related as well! We think about everything.
What if we could store our thoughts and ideas somewhere for future reference? Communicate and build on them? That is what “Asyra” will help us do.
The End Goal
The the end goal for this project will be constantly moving as we modify our ideas of successful outcome. A successful outcome at this stage would be to see if we can cultivate more interesting chatbot responses for our users. Responses that represent Asycd’s motivation for progress and interest in various fields.
Other objectives include:
Improving the contextual responses. More data is always good
Creating a human like version of a chatbot trained on my thoughts
We can also use our NLP expertise to analyse the sentiment of the new responses compared to the old responses. This experiment will bring about all sorts of opportunities to learn how best to optimize generative AI models.
Should I be Concerned?
Not for one moment while I have been thinking about doing this have I had any sort of sense of risk being missed. I understand that it will be potentially detrimental if I reveal anything personal like home address, financial information or similar. Other than that, I think I will be fine and so will Asycd.
Tripe AI movies tell me that this will most certainly lead to my demise but considering the fact the AI is not tangible and able to kill me, I will definitely be fine on that front.
How Will it Work?
The Database of Thoughts and Ideas
As illustrated above, the retrieval process will utilize the embeddings approach to find the relevant ideas from the vector database and inject them into the user’s prompt before delivering that prompt to the LLM.
If you are not sure how the embeddings process works, in short it converts your data to a vector and that is stored in a vector database. When you submit a query, that query is also converted to a vector and to find the most relevant text to your query, a similarity search is done(Cosine Similarity search is the most “common”).
So queries made to my chatbot will be matched semantically with relevant thoughts or ideas from the database.
Daily Thought Ingestion
Since I write everyday anyways it will just be a matter of feeding text data into the models sources for training.
Privacy + Safety Disclosure
Asycd is constantly exploring new model training methods to provide more context. While we understand the importance of data privacy, we are willing to push the boundaries and experiment with sensitive data to enhance our AI’s understanding. However, we prioritize the security and confidentiality of personal information and ensure that no sensitive data is compromised. This includes people involved with Asycd and outside of it. We aim for this to be a safe exploration.
Key Safety Features:
Mild Content Censorship (Built-In)
Manual Prompt Reviewal
Name Blocker (except my own)
These are a few of measures in place to make this process as safe as possible.
Test it Now!
The chatbot is available to test right now!
No two responses will be the same with Asyra since the training data is constantly changing and being added to.
Explore the different ideas and concepts to do with Asycd by clicking the bubble icon in the bottom right hand corner
If you would like to reach out in regards to collaboration, a query or anything else, click here.
