November 7, 2024

A New Language

Working with AI is like blending your brain on high. You can work on multiple projects back-to-back-to-back with moderate efficiency and that’s amazing and tiresome at the same time. AI has made me want to build more and so I start a new project every time the opportunity presents itself. My creative mind has become increasingly fractal, spiraling outward in every direction, and to keep up with my plans my brain needs to stay focused for longer.

Add a script here, apply a change here, switch windows, send a request, do this thing like this other thing but with a slight variation.

Doing everything at once the way that modern computers and languages want you to is tiresome but there will be a better way soon. A new language will emerge for AI tool users that’s full of new abbreviations and operations. All of the clutter from english will be swept away and developers will create a spoken language of efficiency that only needs to be understood by machines.

We need to be careful that this language doesn’t degenerate towards gibberish, though. When working with LLMs it’s common to throw grammar to the wind and type in pop-short sentences. “try catch this”. Spelling rarely matters. The same typos have been made a billion times and absorbed into the AI’s processing system so that it doesn’t matter if you type human or hunma. The machine get’s it.

Deepgram for STT, ElevenLabs for TTS

Without having done any research, I was under the impression that ElevenLabs was a direct competitor to Deepgram. That’s not true, at least completely. ElevenLabs creates models for turning text into speech and they are pretty good at it. They have a large library of default voices and provide tools for creating new voices. Deepgram also has a TTS offering like this but you’re limited to using their built-in voices. Deepgram differs in that it has a STT model as well so you could theoretically use Deepgram across your entire voice stack.