It was fun to experiment with, but it’s obvious that 90% of their development effort has been going into censoring their models instead of improving their utility. I saw no tangible improvements over months.
For example, the web browsing extension was released completely broken and then… remained broken. Meanwhile Phind.com has been doing web browsing very well and very fast.
The Wolfram extension was also useless, and in the same time period a Mathematica update was released with a far superior LLM notebook mode that is actually functional.
OpenAI also don’t allow access to their long context length capable models in the Chat web app.
I switched to the API subscription because it is billed based on consumption and I can use the 16K context GPT 3.5 models.
Meanwhile, despite announcements of “general availability” I still can’t access GPT 4 via an API and nobody has access to the 32K context version. That would be truly useful to me and worth paying for… but OpenAI does not want my money.
I guess I’ll just have to wait for Anthropic or Google to make a GPT 4 equivalent AI that I can access programmatically without have to prostrate myself in front of Sam Altman.