Our Early Impressions of Gemini
Our Early Impressions of Gemini
OpenAI now has real competition from Google.
Gemini, powered by Google’s leading LLM Gemini Ultra when using the “Advanced” version, is now available to everyone, marking the first significant competition for OpenAI's GPT-4. We signed up for Gemini Advanced this morning, and while we certainly have not put Gemini Advanced through its paces, we wanted to share our initial impressions. If you’re looking for a more in-depth view, we’d suggest Ethan Mollick’s post from this morning.
Our primary observation: Gemini Advanced is a genuine competitor to GPT-4. When Google first released benchmarks that showed Gemini as comparable to GPT-4, we had questions about how that comparison would hold up in practice. At first blush, it holds up quite well. The output quality of both Gemini Advanced and ChatGPT Plus appears similar, even if there are stylistic differences. Gemini Advanced has the edge in speed, but our prior experiences with ChatGPT lead us to believe speed may fluctuate over time.
With that said, there additional thoughts and observations from our first hours with Gemini Advanced that we want to share.
The integration with all things Google (Gmail, Drive, Search, Maps, etc.) sets it apart. Gemini can pull information from and summarize what’s happening in your inbox, create travel itineraries with Google Maps directions embedded within it, and use Google search to verify facts (though it still isn’t flawless). If Google continues to push for search integration, we may see it take over the space that Perplexity currently holds.
Image generation is solid, beating Dall-E in our view while still falling short of Midjourney’s image quality and flexibility. It handles text inside images well while being more artful and less cartoon-ish than what ChatGPT produces.
The lack of custom instructions and GPTs limits its utility for more complex prompts and use cases at this point. There’s no indication that similar functionality is on the roadmap for Gemini, but we expect Google to have an answer to this down the road.
Many of the same quirks and limitations we’ve grown familiar with through ChatGPT remain true using Gemini. We’ve found the skills and intuition we developed with ChatGPT apply for Gemini, too.
Gemini, even without Advanced, gives everyone another option to access a leading LLM for everyday use. For most, it will give them the first chance to use generative AI that meaningful connects with data and information they work with on a routine basis. And it lets users do so under a well-known brand, Google.
We expect we’ll have more to share on Gemini Advanced in the coming weeks and months. In the meantime, we will begin to use it in parallel with ChatGPT so we can better understand the comparative strengths and weaknesses of the two. We encourage you to do the same.
AI Disclosure: We used generative AI in creating imagery for this post. We also used it selectively as a creator and summarizer of content and as an editor and proofreader.