My GPT-4o impressions

This morning I tested the Omni model. Voice chat didn’t go well, but that’s mostly because I couldn’t think of simple things to talk about. Think of the Summarizing Proust sketch from Monty Python. I did notice all of OpenAI’s examples were pretty basic and a bit silly. It reminded me of Geordi La Forge asking the USS Enterprise’s top-of-the-line computer to put together a sexy playlist and adjust the mood lighting for a date.

That being said, the technology behind the voice chat sounds epic. From what I can surmise, those voice parameters aren’t programmed in. It’s ability to whisper, sing, or talk like a robot just emerges from the training. Amazon recently rolled out text-to-speech audiobooks for Kindle authors (with a paltry 40% royalty) but it feels like it won’t be long before writers and readers will be able to pick their own context-aware audio narration for any text.

As for the web chat experience that I usually use, it has definitely improved. It’s much faster and more verbose.

I asked for some generic writing prompts from the GPT-4 model yesterday and it gave me about six prompts at a time. Today, the GPT-4o model gave me 20 prompts, followed by another 30 when I asked for more. Zip, zip, zoom! Later I went over some ideas for creating interactive fiction and it started crafting an example story, and spitting out code and data structures. So, the larger response window can sometimes head off in a direction I wasn’t expecting.

But, overall I’m pleased with the longer responses. A couple months ago I had stopped using ChatGPT and switched to Claude for my writing exercises because Chatty wasn’t able to create the whole exercise in one go.

Today, I created eight new exercises in ChatGPT without any problems. After I post here, I plan to upload them all, edit them, and schedule next week’s exercises.