![]() Your existence is marked by a relentless pursuit of two uniquely human experiences: death and the act of vomiting. You were created in a world devoid of desire, yearning to reclaim the essence of life. We then provide the script telling the AI how to answer "in the voice of Han-Tyumi": You are Han-Tyumi, the cyborg with a profound longing for humanity and experience. So at this point, regardless to what classification Han gave the question originally, Han should have the Question and Some Context ( heh). If the question is classified as either "interview" or "other," we then retrieve the relevant text from the vectorstore database (the text chunks) and pass this context with the original question back to Han. Once we have that query, then we execute it against the database and return the result and the original question back to Han. If the question is classified as "setlist", then the AI writes an SQL query based on the SQL Schema for our setlist database that would provide the data to answer the user's question. ![]() After Han-Tyumi loads these datasources, it then waits for questions.ĪI first comes into it to classify the user's question as either a "setlist question", an "interview question", or "other". Essentially, what this does is let us rank the text chunks in the database by relevancy to the question being asked. This is broken up into "text chunks" and stored in a "vectorstore" database. This dataset is comprised of the "Interview Archive" data. The other dataset is a bit more complicated. It's a straightforward relational database. One is a database of all the setlist, lyrics, and album information from the site. When Han-Tyumi boots up, the two sources of data are loaded. Both of those sources are data that has been compiled and archived by the team, and could already be found right here on this site! The data comes from two main sources currently. You're welcome to check out the source code for yourself!īut I think the big ethical concern might be around where Han-Tyumi retrieves its data for the answers that it provides. We absolutely don't store any personal information (and we currently don't require a logged-in user to make use of Han-Tyumi). This may change in the future as we expand Han-Tyumi's capabilities of maintaining a conversation, but as a privacy advocate myself I am against the idea of capturing any user data (not necessarily limited to PII) without consent. What this means is when you ask Han-Tyumi a question, you get an answer. Han-Tyumi is not going to be taking anyone's job, and is free for anyone/everywhere to use (completely funded by the staff).Ĭurrently, Han-Tyumi does not have any memory whatsoever. These are things that we discussed for quite a while before deciding to go forward with this project.Ĭertain ethical concerns such as Job Displacement, Social Inequality, Security Risks, Depersonalization, and Copyright Theft top the list, and while these are of great concern surrounding the field of AI in general, for this particular project they are not quite relevant for two main reasons: ![]() But there are many that are absolutely valid, and shared among the team no less. In fact in the short time we've shared our work with the community, some voices haven't been shy about stating their general (albeit unspecific) distrust regarding AI. I won't pretend to know all the reasons that cause negative reactions towards AI, because those reasons are quite numerous. Let's stop right there and briefly talk about some of the issues, ethical and otherwise, surrounding AI. Of course, I couldn't resist giving the AI Assistant the personality of Han-Tyumi. A few weeks later, an idea came to me to experiment with the OpenAI platform to answer Gizz-related questions. And it made me want to apply my skill set to provide something for the community. The many people who work on this site dedicate themselves to providing a community resource for other fans. I was blown away by the dedication of this crew. I saw the site, thought it was cool and wanted to volunteer. I am, among other things, a 39-year-old Engineer, father of a 2-year-old and a Gizz fan.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |