The talk of this week has been the slew of announcements from Open AI on its first DevDay, in which Open AI made a big effort to enroll more developers into its platform and create a sustainable and irreplaceable ecosystem, modelling in some ways what Steve Jobs did in creating the App Store ecosystem for the iPhone.
In this short post, I will review what I found most exciting and relevant for research and development in robotics. While a lot of airtime was around making GPT models more useful and relevant, the most promising announcement was the GPT-4 Turbo With Vision.
GPT-4 Turbo With Vision
The latest multi-modal (A multi-model model is one that can consume both text and images) that Open AI offers (only to paying customers) is GPT-4. Open AI is now making available a GPT-4 model with a sequence length of 128,000 tokens trained on data as recent as April 2023.
The sequence length implies the number of input tokens that a model can consider to generate an output…