Is Google Gemini Capable of Planning?
It’s mostly a matter of degree.
In just a few days since the public launch of Gemini, Google’s new AI platform (or rather, family of platforms, because there are three flavors, more on this below), much has been said about it, from “ChatGPT killer” to “No innovation there.”
Many of those headlines sound outrageous on purpose because they are intended to get eyeballs.
But we wonder, is there real meat in Gemini, or is it just a desperate PR stunt trying to present Google as a “cutting-edge” company?
I made some experiments myself, keeping one specific aspect of Gemini in mind: its planning capabilities.
There are many other aspects of Gemini that I haven’t tested so far, like its multimodal capabilities. But planning is essential for reasons I explain in the following.
Planning in AI
What has been discussed in the last year is why conversational AI systems like ChatGPT or Bard have had so much trouble when it comes to planning sequences of steps to achieve a desired state.
I’m not talking here about planning in AI in general because, in traditional AI (not deep-learning LLMs), much progress has been made since many years ago.
