We tested Gemini Pro on Google Bard, a worthy pretender to the throne

Google recently unveiled its riskiest bet in the field of generative AI, a bet that directly attacks the domination of Open AI and its ChatGPT: Gemini is its most ambitious model, a model suitable for any device, including Android. And this can now be proven: Gemini Pro is active on Google Bard if a very specific process is followed. We tried.

ChatGPT’s dominance made Google nervous, it’s not for nothing that Open AI overtook the search engine by taking advantage of its training and machine learning tools. It was time to turn the situation around, Gemini is the multimodal model that has come to dominate them all. Actually, there are three models: Nano, intended for devices like Android; Pro, for human interaction tools; and the Ultra, an advanced model with which Google intends to revolutionize segments from research to development. And Gemini Pro can now be tested.

Google Bard finally lives up to what Google deserved

I’ve been testing Bard since Google launched it in private beta, and the truth is that until now it hadn’t evolved much: I recently compared it to ChatGPT and it still failed miserably. The AI ​​didn’t think creatively enough, data was often incorrect, and the code developments did not seem satisfactory to me. So, with the arrival of the Gemini Pro, it became clear to me that I needed to test it to see if the new model had addressed the flaws of its predecessor.

To begin with, it is not very easy to test the new model because It is only available in English. However, it is about applying a few steps and knowing how to get along with the language:

• Log in to your Google account and go to “Personal information”.
• Go down to “General Web Preferences” and enter “Language”.
• Select American English as your preferred language. Alternatively, you can access this link by avoiding the previous steps.
• Now you need a VPN to fake a connection from the United States. I chose NordVPN.
• Load a US server and navigate to the Bard page. It should appear in English to tell you that you are using the Gemini Pro model.

Once Google Bard was active in my Android web browser, and after making sure Gemini Pro was working, I asked it various questions related to different domains. Direct questions about historical data, mathematical calculations, problem solving, or creative reasoning around philosophical problems. I noticed a clear evolution in behavior, both in terms of satisfactory results and in the way of reasoning.. Although I saw some difficulties when following the thread of the conversation when I returned to a concept already mentioned, which does not happen with ChatGPT.

Google rightly assures that Gemini Pro is superior to GPT-3.5, at least in terms of neural performance tests. And the truth is that They are on a very similar level, at least according to the tests I’ve done. Of course, Google Bard has nothing to do against GPT-4, at least until Gemini Ultra and its multimodal operation are available; something that will happen in early 2024.

I really miss being able to ask for images, for example, also for more creative reasoning: Gemini Pro is a bit cold, it is difficult for it to deviate from logic; Although he surprised me in more than one response, like when I asked him for a Christmas card and he wrote it to me in Spanish because he felt I was in Europe (even though Nord VPN was connected to the United States). It appears faster than its opponent Open AI, the response speed is at an excellent level.

Currently in limited testing in the United States

I tested the new Bard, both from the smartphone and from the computer, and the result was satisfactory in both cases for all the tests: where Bard was wrecked before now seems much more reliable. This is not the best way to ask questions, but little by little we are getting closer to what a generative AI should be in terms of consultation. The multimedia generation would be lacking.

Greater reliability, better reasoned responses, increased scope of action, notable speed and even a certain approach to human logical thinking, always filling the gaps posed by training-based Artificial Intelligence: Google does its homework, even if it is (much) late. I can’t wait to see what Gemini Ultra has to say, hopefully we’ll see it early this year with an experimental version for Google Bard.

Gemini Pro can currently be tested, just change some Google account settings and use tools like Nord VPN.

Pictures | DALL-E 3 on ChatGPT

In Xataka Android | How to get the news about ChatGPT for Android before anyone else