So I spent about 3 hours working with an AI to develop an AI prompt optimized for minimal execution time that would tell the AI to identify product data, product claims, and scientific evidence on a few hundred products, cross reference those data and rank products by likelihood of working as claimed.
We're not there yet, but I really like that on our trial run, the same AI that had helped me write the prompt, upon realizing that it was the chosen executor of that prompt, began it's response with, "OK, This is a massive undertaking..."
Unus Nemo likes this.
reshared this
Unus Nemo
in reply to Blaise • •@Blaise
Are you using a custom LLM or are you using one out of the box? By custom I mean one you built yourself not just one you trained. The results are going to be very interesting as product claims can rarely be accounted for in product data. Marketers have a way of ... well lying is the only appropriate term I can think of. I would love to hear more about this project as it continues. Please keep posting to keep us up to date.
Blaise
in reply to Unus Nemo • •@Unus Nemo No, I'm experimenting to see how much useful work you can get out of a free, public AI service. I generally start with Grok, then attempt to reproduce on ChatGPT and Claude.
My whole project may come to nothing, in the end, because the damned things keep making up some percentage of the data, even if you spoonfeed all the research materials to it and specifically ask it to use only, and all of the data provided. And they lie about what and why they did what they did at least 80% of the time...
I'm at best only about 80% there. I'm not sure my theory of getting the AI to help me write the prompt will turn out to be helpful...
Unus Nemo likes this.