Google stands up to Open AI with a generative AI chatbot that also costs money: Gemini Advanced is the answer to ChatGPT Plus. Both have enormous potential, boast the most advanced language models and promise versatile operation with ability to answer most questions. So it was time to put them to the test, to see how Google’s new thing stands up.
The range of generative AIs that we have at the touch of a screen is getting bigger and bigger once the key technology companies join in to compete with them. ChatGPT, from Open AI. And Google could not be missing from this competition, not in vain is a pioneer in training, modeling and execution tools. With all that experience, Google Bard first came out, a chatbot similar to ChatGPT, then it evolved into Gemini. With a paid variant that I have been testing since it came out.
Gemini Advanced, a paid AI that adds more creativity as its main value
Google’s Generative Artificial Intelligence has become very confusing once the company buried Bard to exchange it for Gemini. It currently offers its models as a chatbot in the Gemini page: If you enter for free you will be using Gemini Pro. Do you want the most powerful and capable of all? You must pay for Gemini Ultra, it is within Gemini Advanced. Well, you can take advantage of the initial offer and try it for free.
Let’s get into flour. As soon as it was made public I went to Google One and upgraded the subscription to have access to Gemini Advanced. The premium chatbot loads automatically if the user has subscribed, although this does not imply an addition of options to what was until now Google Bard: all answers in Spanish are text. Gemini can offer images, but only taken from the web (in English and with a US VPN it does generate graphic content).
Google’s AI Premium subscription does not add options to what you can do for free with basic Gemini, the chatbot that is based on the Gemini Pro model. So, why pay for something that you already receive for free? The key lies in the creative capacity and the AI’s abilities to adapt to different tones of conversation.
After the numerous tests I have carried out, after numerous threads, requests and attempts to get Gemini mad, I can confirm the fact that Advanced is much better in the facets of creativity, naturalness, humor and proactivity. Even so, it continues to fall into errors that carry over from the original Google Bard, a chatbot to which you could pass any false data as real.
Gemini Advanced is not at all reliable with the data it offers (no AI is, but Google’s raises the bar), it does not fully capture the tone of the requests, insists on being proactive when no one asks her to, usually ignores conventions and often responds with a refusal to a request to follow it in his own way with the next step. It has baffled me on more than one occasion, sometimes for the better.
Gemini Advanced says more than what is asked of it, does not fully understand all the nuances of an order and tends to give numerous erroneous and unverified data
I put Gemini Advanced face to face with ChatGPT Plus, a tool that I have also been paying for since it came out. Roughly speaking I confirmed that Open AI’s chatbot is still much better at human interactions than its opponent, it is more versatile (it can create images, something that Google only offers in English and with a VPN from the United States) and it captures instructions on the fly to react accordingly. If I pay for an AI, I am very clear that ChatGPT is a much better option, at least for now. Although Gemini gives you 2TB of Google storage and many more benefits, including Google’s VPN.
«But what are you telling me, Gemini Advanced!»
Google’s chatbot is faster than ChatGPT and tends to expand its responses to a greater extent, often adding important context to what is asked. One of the options that I like the most is that Gemini allows you to vary the tone of each response by clicking on the edit icon, a tone that is very well chosen. It allows you to apply it to the entire conversation and tends to correctly remember the previous premises, although sometimes you get confused and respond as if you had no prior background (you can also cut off a response if you detect that you have gone too far). This does not happen to ChatGPT.
While in an impersonal conversation, Gemini Advanced decided to make a joke about Catalans without meaning to it, from there he underlined that shows remarkable proactivity.
It is often difficult for the AI to understand a very clear instruction. For example, spin a story where each of us assumes a role to see how far the creativity of the other goes. I tried a typical writing exercise that Gemini Advanced was unable to carry out, even though I gave the request a thousand thoughts. Nothing.
However, ChatGPT Plus caught it the first time. And I was surprised by how well he got into the role.
In general, he doesn’t like short answers, Gemini Advanced is like that person who knows Wikipedia as if he had written it in his own handwriting and likes the way it is recited. Often with wit.
And when it comes to creating content? I found it to be a very good tool: it perfectly captures the nuances and knows how to explain itself naturally regardless of the format requested. The problem is the data, it is easy for him to believe what you tell him, at least in information that is not too relevant. I couldn’t make him believe that I was the President of Spain. However, I did run my own company.
Creating code Google Bard had its pluses and minuses. So it was time to see how Gemini Advanced performed. I started with a bookmarklet to automatically download all the images on a page, but I couldn’t get it to work (ChatGPT came up with something decent the first time). Then I tried creating a website counter to analyze the number of characters, words, spacing and code on each page. Through several attempts I managed to get something close, but it was not what I was looking for. I doubt it performs well in complex developments, although I don’t have the level to verify it either.
Gemini Advanced does not create images as is, although it can if the chatbot is used with a VPN connected to the United States and given instructions in English. This has nothing to do with the subscription or the language model, since Gemini uses Google’s Image 2 model for this. It is also included in the free version of Gemini.
Conclusion: Gemini Advanced is still too green to pay more for it
After my hours fighting and laughing with Gemini Advanced, as well as getting frustrated with some requests that there was no way to complete, I can conclude that The jump from Gemini Pro to Gemini Ultra is palpable, is an improvement that can be seen in the naturalness and creativity of the chatbot. Even so, it continues to maintain absurd errors, such as the amount of data that is invented, how easy it is to deceive it and the fact that Google does not incorporate really new functions to those who intend to pay for its Artificial Intelligence.
Compared to ChatGPT, it is very clear to me that Open AI’s product is much more polished than Google’s, it works better in general terms, adds more possibilities (why is image generation in Spanish not included?) and is shown more resolved. Also It is slower, limited to 40 responses every three hours (in the most advanced model, GPT-4) and does not offer the 2 TB of storage that Gemini Advanced includes, which also enter the fight to take first place on the podium.
After my tests, and if I only opted for generative AI, ChatGPT Plus seems like a much better investment to me. I am sure that Gemini Ultra 1.0, the version that includes Gemini Advanced, has a very large room for improvement, but the feeling it currently leaves is that of being very green. And in a product that is supposed to be premium, it does not seem acceptable to me.
Cover image | Ivan Linares
In Xataka Mobile | Google Bard is now Gemini and comes with an app to stand up to ChatGPT