I use X (Twitter) to get most of the recent releases, news about AI models, etc. Then, if something sounds interesting and might be related to something I am building, I will try to test it and see how it works. This was the case with https://huggingface.co/microsoft/VibeVoice-Realtime-0.5B
So here is the exact prompt for codex (in VSC):
look at https://huggingface.co/microsoft/VibeVoice-Realtime-0.5B and create the sh to install requirements, and create an web page that uses the model to speak, also in that web page, connect with a local llm, to create a chat interaction, and test the streaming oputput from the llm, i will past the llm endpoints that will come from lm studio, and the web ap should have a chat interface, that after i enter the text will send to the local llm, and on on streaming it will play triugh the tts, also have the options for voices etc. Build it in python
Now this will produce a simple path for me to test the new shiny thing. Of course, this is not the end; it’s just the starting point. From here, I have to resolve installation issues, change how it works or looks, etc. But I say it was done in one shoot at about 95% level. Yes, this is simple stuff, and more complex things require a lot more steps, but nevertheless, testing new cutting-edge research just got easier, and you can add them to your tools, make better-informed decisions about what is available, and how it works, etc., so that when you actually start engineering a product, you have a better starting point.
Far from this being any prompt library thingy, do note my spelling mistakes as this was written at almost the speed of thought, I am hoping this just encourages others to test new stuff, with their own preferred setup.
