Natural Language Assistants

Posted on 9 October 2023

I recently created a demo Alexa skill that allows me to talk to the Claude AI. (The code follows alexa-gpt but switching out the OpenAI API call with the Claude API).

It was an interesting experiment and it really revealed to me just how nice it was to talk to something that better understood what I was saying and was able to reply in a more human-like way.

Previously Alexa would reply "Hmm I don't know that one" or "Here's what I found" (coupled with an unhelpful reference to an unrelated website). This frustrated me as a user and resulted in me using the device less. I felt I had to change how I interacted with the device to get it to work. I remember one time I said "Open Spotify" for it to rudely reply that I can open Spotify with the command "Play Spotify"! It knew I what I wanted but I apparently didn't speak the right incantation.

With the test skill using a LLM however the results were much better. Whilst the LLM cannot search the web or look up live data it can answer more nuanced questions. I personally find LLMs useful as a soundboard for ideas and to run through things prior to implementation. For example I may ask what it thought of one architectural design over another for a certain implementation. Weighing up the pros and cons of both. Speaking to it rather than typing also meant that I could chat with it whilst doing other things.

I found this very powerful and I highly suspect assistants to start to use generative AI to provide better responses to questions and to understand the users intents better. It will be interesting to see how Alexa and other tools evolve in this space. I would love to have a more natural conversations with Alexa.

I also envisage that assistants will become more personalised (and potentially more characteristic). With generative AI it's possible to have responses have personality and flair. Assistants could be tailored to the individual to provide a nicer experience. We are already starting to see Meta add personality to its AI chatbots. It's only a matter of time before speech-based assistants see the same evolution.

While there are certainly risks and challenges with more human-like AI assistants, I believe the overall impact will be positive if done thoughtfully. More natural and engaging assistants can help people in many ways and I'm eager to see how developers balance utility and personality as this technology continues maturing. One thing is clear - the era of robotic commands is drawing to a close. The future is a real conversation.