Skip to content

Commit

Permalink
tweaks
Browse files Browse the repository at this point in the history
  • Loading branch information
Kayce Basques committed Oct 20, 2024
1 parent 46704d2 commit a06be1e
Showing 1 changed file with 12 additions and 28 deletions.
40 changes: 12 additions & 28 deletions data/embeddings.rst
Original file line number Diff line number Diff line change
Expand Up @@ -245,27 +245,20 @@ prolifically?
Applications
------------

I wanted to say more here, but then I realized that my employer probably
wouldn't appreciate me throwing around potential state-of-the-art ideas
willy nilly. (Of course I could very well be wrong about all this and
my ideas could amount to nothing. It would be very nice however to keep my
job, especially because my wife and I are preparing for our first baby.
Better safe than sorry.)

So all I can provide is one basic example and a hint about where
to focus attention further.
I could tell you exactly how I think we can advance the state of the art
in technical writing with embeddings, but where's the fun in that?
Let's just cover a basic example to put the ideas into practice and then
wrap up this post.

Related pages
=============

Some docs sites have a recommendation system that makes you aware of other
docs that may be relevant for your needs. Basically, they look at whatever
page you're currently on, find other pages related to this one, and suggest
you to check out those other pages. Previously this was usually implemented
with keywords or user data. Both approaches are a lot of work and are pretty
hard to get right. Embeddings provide a new way to provide this feature at
probably a fraction of the cost (in terms of time and energy spent building and
maintaining the feature):
page you're currently on, find other pages related to this one, and recommend
that you also check out those other pages. Embeddings provide a new way to
support this feature, probably at a fraction of a cost of the status quo
methods. Here's how it works:

1. Generate an embedding for each page on your docs site.
2. For each page, compare its embedding against all other page embeddings
Expand All @@ -284,27 +277,18 @@ See :ref:`embeddings-appendix-implementation` and

See `Related content using embeddings`_ for another example of this approach.

A hint about "state of the art" potential
=========================================

Throughout this post I have been saying how embeddings enable us to compare
one piece of **text** against any other piece of **text**. I did not limit
the description to just **documentation**. "Documentation" is a subset of
"text". The work of technical writers involves many types of texts.

-------------
Parting words
-------------

Three years ago, if you had asked me what 768-dimensional space is,
I would have told you that it's just some abstract concept for physicists
and mathematicians. But now, embeddings have given me the motivation to
grok 768-dimensional space at a much deeper level. Most importantly,
the concept is directly useful in my own life. I think that's beautiful
and exciting.
and mathematicians. Embeddings gave me a reason to think about this idea
more deeply, and actually apply it to my own work. I think that's
beautiful.

Order-of-magnitude improvements in our ability to maintain our docs
may very well still possible after all… perhaps we just needed
may very well still be possible after all… perhaps we just need
an order-of-magnitude-more dimensions!!

.. _embeddings-appendix:
Expand Down

0 comments on commit a06be1e

Please sign in to comment.