AI source tagging

daviddelven

Does it make sense (anymore?) to tag the AI feedback during the Input/Capture stage, distinguishing the AI-provided info from our own or third-party sources? Or do you treat it as a regular web source?

Sascha

I'd never rely on AI as a source. AI lied to me a whole lot and I don't trust it.

But I'd always make sure to reference the source.

Will

@daviddelven, thanks for bringing up this topic. I hadn't thought through this topic before I started writing a reply. As usual, I don't know what I think until I start writing.

Yes, it makes sense to 'tag' AI-related feedback, just as providing any 'feedback' attribution makes sense. Every reference should be traceable to its source. AI-provided info is a third-party source.

I'm unsure what you mean by "do you treat it as a regular web source?" For me, web sources are listed as footnotes as Markdown links in the Works Cited section at the end of notes. I've not looked into it, but getting a link to specific AI-related feedback and promit would be next to impossible.

I'd suggest including the prompt along with the feedback. Both have value.

I'm still experimenting with providing attribution when including AI-provided info in a note. But I think it is not much different than other sources. I think I'll use the quote format and a one or two-word key for the AI tool used. I don't plan on including too much AI-provided info in my notes.

Here is an example of how I might do this.

[Prompt]
You play the part of a seasoned 30-year motivational speaker. There are lurkers on the forum that I'd like to motivate to surface and participate in discussions. What five motivational sentences can I use to address them?


  [OpenAI GPT] 

These are all small notes. I find a description like this provides both attribution and context. Here are a couple of examples of attributions to AI-provided info. These sentences are at the top of the note.

1. [[Task Serialization ChatGPT3 202212101448]] The first paragraph is refactored by me. The rest is directly from the ChatBot. Grammarly Plagiarism found 12% of the first paragraph questionable, so I rewrote it.
2. [[Inspiration for Incorporating Specific Emotions in Writing 202303191143]] Most of this note has been written by OpenAI's ChatGPT in response to my prompt. This will help me write the final paper in ENGL501 about spoons.
3. [[Inspiration for Specific Writing Themes 202303191129]] Most of this note has been written by OpenAI's ChatGPT in response to my prompt. This will help me write the final paper in ENGL501 about spoons. Below is the output for OpenAI's ChatGPT. It is surprisingly helpful. There are several creative tangents I can explore. This has given me the idea that the subject is rich enough for my final project in ENGL501.

Sascha

@Will wrote that:
I'd suggest including the prompt along with the feedback. Both have value.

I second that! A very important reminder.

daviddelven

This is how I keep the original AI prompt.

Another stuff is about how to keep the set of prompts when the AI engine performs following a conversational mode (i.e. chatGPT). The set of prompts that you include in the conversation is, per se, an outcome of the provided feedback from each of the previous prompts on the series.

Sascha

I think there comes a point at which you'd summarize the AI interaction and then work with the result.

ctietze

@Sascha said:
I think there comes a point at which you'd summarize the AI interaction and then work with the result.

With that in mind: a transcript of the chat would be a data point like all the other (interview) transcripts are. So if in doubt, export and save the chat, treat it like any external resource, and then summarize it like with a reference in your notes?

Sascha

@ctietze said:

@Sascha said:
I think there comes a point at which you'd summarize the AI interaction and then work with the result.

With that in mind: a transcript of the chat would be a data point like all the other (interview) transcripts are. So if in doubt, export and save the chat, treat it like any external resource, and then summarize it like with a reference in your notes?

Exactly.

I don't store sources. So, my inclination is to do the same with AI.

daviddelven

I agree to treat the "conversation concept" as a source, but still the "my questions" within that conversation could live out of the source, even a "zettel" could become a single prompt inserted at some point of that Human-AI-human conversation.

I store my sources, all of them. Out of the slipbox, but totally accessible. They are part also of the feedback of my inquiries, thus somehow the history of what brought me to some outcome.

I guess you meant storing them in the "slipbox".

Sascha

I guess you meant storing them in the "slipbox".

No, I don't store sources unless I go back to them frequently.