Forum Replies Created
-
Posts
-
<div class=”bsp-mentions”>@jtkolarik, I would assume that since the nature of the project is to collect and Dave wisdom that there would be some value in recording answers through all of the vicissitudes of life. Our perspective and our opinions can change over time, and so there’s probably no perfect moment to record any given answer, but let’s say you run through all thousand questions in a year or two. Hit them again later on in life, well past the point where you’ve forgotten your original answers, and see if you come up with something better than before. I’m sure an AI built with your collected wisdom over time would only improve with more data versus a limited one based off of how you felt you were in an optimal state to answer.
</div>
<div>If I might be mistaken on this thought, please let me know what you think.</div>@jthorn, I’ve been introducing it by sharing the general idea, and then making an example of how some families would keep a private library. I ask them to imagine if we had our own private library filled with all of the wisdom of our family that we could refer back to, except instead of a shelf full of books, you could interact with a language model. So far I’ve got a few family members on board, and I’ve already bought a couple recorders to hand out.
I’ve picked up a recommended audio recorder (Sony ICD-UX570), and I’ve also been using this software to transcribe the audio: MacWhisper
MacWhisper is an independent app that uses OpenAI’s Whisper ai voice recognition model, and for around $20 on the pro version, it uses their large voice recognition model. It transcribes audio slowly, and with relatively high fidelity. Unfortunately, there is no speaker diarisation, but if you are going through the wisdom savers solo, that’s not an issue.
I have been using Macwhisper to transcribe test audio and save the text files for gpt4all’s local docs. I’m putting this out here to see if anyone else has or will try it. I’m curious to see if the wisdom savers can be built on incrementally over time, but as the audio is transcribed, we might be able to make use of these words on a private LLM.
Right now, the local docs plugin isn’t particularly amazing when compared to the chatgpt pdf plugins, but the idea of taking a person’s wisdom and being able to chat with their ideas practically instantly is an intriguing one. I hope to see some future posts of others trying to do the same, and I’m curious to see what gets better results.
