Next Gen OpenAI Models: From Iterative Improvements to a Capability Jump

Developing with AI is a tricky endeavor:
Sometimes our work is outright replaced by new model capabilities.

In a recent interview at the ‘AI for Good’ conference, Sam Altman, CEO of OpenAI, made a statement about what to expect from the next generation flagship model. He anticipates that the new model will significantly improve in some areas but not as much in others.

This statement differs from what Sam Altman has previously communicated about what we can expect from new OpenAI releases. Following GPT-4, the outlook was to expect gradual improvements over time. And I can say that this has mostly been true, especially for large language models and the accompanying embedding models.

If we can now expect a jump in capabilities:
In which areas will it be?
And which areas will progress more slowly?

I think this is an important question for developers, as we should consider if our efforts may be replaced outright by new model capabilities.