Dictionaries and Rich Hover: Is AI the Answer?

Long ago, @Paul_Danyliuk gave me a number of ideas concerning ways to create rich hover elements in text. It led to this Rich Hover demo document.

Recently, @June_Parent1 raised a question about Sharing Terminology & Definitions. I enjoyed writing the response enough that I decided to make it a post of its own. This topic needs more visibility not because I believe it should be a new Coda feature, but because it shows (again) why access to text as first class objects is so important.

The fact that we cannot retrieve the fifth word in paragraph three through a formula is a tragedy. The fact that we can’t replace the fifth word in paragraph three through a formula is a tragedy.

With my disappointment in full view, lets explore how we might approach the creation of an internal dictionary of terms in a practical sense. There may be other approaches, so please share.

Page Embedding - this makes it possible to easily embed another Coda page or document into other documents. As such, you could create an internal dictionary and as a matter of practice, simply include it in documents where it may matter. Recent feature additions to page embeds allow you to establish the bast approach.

Rich Hover - this is an approach that makes it simple work to create terms in a table and then reference those terms with added in-line visibility. It’s a very powerful approach for drawing attention and instant lookups to in-line texts.

AI Embeddings - this approach, unfortunately, cannot be implemented unless the text exists in a table. It will not work for text on the canvas in a document because we have no formulaic or API access to text. I truly hope the Codans make this possible.

Creating a universal ability for Coda documents to map documented terms, and even similar terms to a dictionary is possible. It requires the ability to (i) extract all key words from the text, and transform each of them into embedding vectors, and (ii) linking the keywords to the dictionary (which is also vectorized). Terms that have a high similarity score will then be transformed into rich hover formulas as demonstrated earlier.

I’m waiting for someone to wave some cash in front of me to implement this.

Coda’s AI Fumble (purely my opinion)

In a rush to provide AI features, I believe Coda took the fastest pathway to show it had AI chops. It wanted the me-too satisfaction of the AI hype-cycle. The reality is that most of the AI features recently introduced by Coda were being used by my team At Stream It over a year ago. Packs made it possible for me to integrate OpenAI’s APIs way back to GPT-2 in 2021.

The AI hype-cycle I’m referring to is ChatGPT completions. This is the laziest form of AI, typically bolted onto an existing product in a hasty attempt to get that checkmark for Gartner and unwitting enterprise customers. It is the knee-jerk of AI, a plague that still ripples through software companies desperate to have an AI story.

Coda’s infusion of GPT-like completions is sadly representative of a broad remedy for laziness which has two dimensions - it’s the corner to cut if all you care about is being an “AI company”, and it’s exactly the right feature for customers who are attracted to the big productivity mirage.

AI, specifically AGI (artificial general intelligence), has its place, and there are vast opportunities to employ it for great user benefit. Innovators often mistake newfound AI capabilities as features.

Artificial general intelligence (AGI) straightens the line between what is known and what we need to know while adding our own context.

This is exactly the nature of @June_Parent1’s suggestion - a feature that straightens the line between workers and what they need to understand. This is the promise of AI, features that magically transform information into smarter experiences.

AI is not a feature; it is a UI that manufactures very powerful benefits for workers.

Coda’s AI release is nice, it has some benefit, but mostly in limited contexts and for a narrow slice of users. For every user who needs to create marketing copy, there are a hundred who don’t. The big wins, hyper-productivity, and an increasingly smarter workforce is where Coda should place AI bets. This is what you’re asking for and this is what they should make possible.