As many of you who know me personally know, I try to stay on the forefront of technology. Rarely am I as impressed as I have been with Claude 3.5 Sonnet. I've had a coding project that I kept putting off since around Easter, and while I had most of the pieces figured out, I didn't have them put together. So tonight, since I just got some more consulting business on the AI side, I sat down with the new Sonnet model and gave it two of the project files. The first and larger one was the speech to text part, where I had identified the place where I could inject the Anthropic API calls for talking with Claude. The second was the conversation code I had written with the API calls. The thing is, I used a fake loop in that code, and it wouldn't actually be a real conversation if I used that code. So, what did I do? I just told Claude 3.5 Sonnet to merge them, keep most of the code the same, rewrite the fake loop to be a real one, and explain what it did. So it proceeded to do exactly what I wanted on the first try, in about 30 seconds. Oh, and it saved tokens by saying in the comments where to copy the boilerplate code from. Not to mention saving me mental stress. I still can't believe it worked on the first try. Now I was able to use Piper TTS to finish making a full speech-to-text-to-AI-to-speech program for the robot. This may actually warrant our first big Youtube video!

Leave a comment