ExtWis: Using AI to Extract Wisdom from Any Text

Member Only Content

Concept: AISimilation of Concepts

If you’re like me you are a voracious consumer of information. Wait, this is
a member post, so of course you’re like this.

The problem

My Our problem is that we
are tracking somewhere between half a dozen and a few dozen input sources at
a given time.

• UL

• Galloway

• Huberman

• Lex

• Sam Harris

• My First Million (I’m currently behind)

• (many others)

And more generally…

• Podcasts

• YouTube Channels

• Books

• Articles

• RSS feeds

• Websites

• Etc.

It’s way too much.

Frustration

My greatest frustration with this isn’t even that I can’t listen to
everything.

My bigger problem is that when I do listen to something I am often
blown away at various points.

• Holy crap that’s awesome! (but I’m driving)

• Oh man I have to remember that (but I’m walking)

• Wow, that’s just like this other idea (but I’m exercising)

Not only that, but even if I were sitting there dedicated to listening to
it, it would take forever to take meticulous notes for every piece of
content I watched.

The ideal format

Just a related AI image that I couldn’t bear to waste

I guess if I were to do that, what would I take notes on?

I think I’d want to know:

• I guess I’d want a summary

• Then maybe like the most interesting ideas?

• Probably the quotes from the text that were really cool

• Any facts that were mentioned?

• Any prior work?

• Any books or podcasts or papers mentioned in the content or
  conversation?

• And I guess any recommendations for what I can do with the knowledge

Enter extwis

How I imagine extwiz working in my head

So that’s what this new project does. I say project but it’s really just a
prompt that I spent a LOT of time on.

Check out this output from
the video
above.

The output from running extwis on the video above.

Look at this thing!

Here’s what it extracted:

• Summary

• Ideas

• Quotes

• Facts

• Prior Work

• Resources

• Recommendations

This is very close to exactly
what I would have written down if I watched the whole thing at 1x and
took meticulous notes. Which, by the way, I would have had to pause the video for.

In other words, we went from:

1. 1x time of video

2. Plus pause time

3. Plus thinking time

4. Plus writing time

…to basically 10 seconds!

I’ve done this for Huberman videos, Lex interviews, blog posts.

It works on ANY TEXT!

It’s fucking WISDOM EXTRACTION!

What about books?

I’m already excited enough about this, as you can clearly see, but there are
some limitations. First of all, size. I have access to a lot of GPT-4
context, but even it can’t fit in a full 3-hour conversation.

And the killer app will be entire books.

I will not sleep until:

1. A book drops at 12:01 a.m. Monday morning.

2. By 12:02 my AI assistant has purchased the book, extracted it as text,
   parsed it for wisdom, and asked me how I’d like to consume it

WANT.

NEED.

The prompt

Anyway, I am so unbelievably happy about this. It’s GODLIKE power at our
hands now.

Hmm. Cool video. Boom. Slurp. Yum. I know Kung Fu.

And since you’re all members here, I’m going to share the prompt with you.
Please keep it to yourself. I am still working on it and may release it
later, but for now it’s going to just be a member thing.

A work in progress, but already works really well

By the way, the tool I’m using for extracting YouTube transcripts is called
YouTube Transcript API, which another creator was nice enough to show me.

youtube-transcript-api

An python API that allows you to get the transcripts/subtitles for a given
YouTube video.

pypi.org/project/youtube-transcript-api

Cool, well I hope you get value from this!

Start a thread in UL Chat if you want to discuss.