On February 22, Nvidia, the main producer of chips for artificial intelligence, presented its financial results. They were so surprising that many interpreted them as a crucial clue that artificial intelligence is serious, not a bubble that can burst at any moment. But many doubts regarding AI remain in the air. Chief among them: which artificial intelligence is better? ChatGPT remains the most promising, but there are two others that also shine brightly: Gemini, from Google, and the arrival of Claude 3 has recently been announced.
Although it is not new AI, among its creators there are people from the team that developed ChatGPT, in its latest update it has taken a great leap forward. So much so that it can be considered a very valid third option in the field of chatbots, or conversational artificial intelligence. We have thoroughly tested the paid versions of ChatGPT, Gemini and Claude 3. Although all three can be used for free with some limitations. To put them to the test we have asked them for rigorous information for an investigation, we have asked them to read texts and summarize them, to respond to emails, to perform calculations on domestic accounting and to explain to us everything they see in an image. Whether this is a photo or a complex diagram.
An advert. These three artificial intelligences occasionally provide erroneous data. Some are even made up. Therefore, the information they provide should be verified. Especially if it is data that we are going to use in some important task. An interesting trick that we have used for this is to compare data in all three. Gemini, for example, is good at surfing the Internet. This is ideal for checking if the books that ChatGPT suggests when requesting a bibliography actually exist on the Internet.
Two clarifications that are important. These artificial intelligences cannot generate images. ChatGPT does this in version 4 using Dall-e. But this is an independent AI created by OpenAI. Gemini has started imaging, but it is still in testing and the feature can only be used in some countries. There are many artificial intelligences capable of generating images. One of the best is Adobe Firefly, which creates them from a text description, as photographs or as illustrations. Firefly will soon be integrated into the free Adobe Express app, which is increasingly popular among content creators.
It should also be clarified that only the creators of ChatGPT have clearly opted for us to be able to use AI on the phone. Something that is surprising especially in the case of Gemini. Google has only launched one application of its AI on Android and with restrictions. It has even been learned that Gemini Nano, a version designed for phones, cannot be installed on the Pixel 8, one of the last two phones launched by the company. A fact that alerts us to how Google is taking somewhat erratic steps in the development of its AI. This is important, since ChatGPT in its mobile version allows us to chat with her at all times. And not just in front of a computer screen. In the way that allows us to talk to her using headphones, we have surprised ourselves by going out to exercise while we listen to the information she gives us about the questions we ask her. A very effective way to learn about any topic that interests us and that we have never experienced before with any technology. It is something similar to being able to interact with a podcast.
Claude 3 is that it surpasses ChatGPT when it comes to being able to provide you with extensive information in a PDF document for you to analyze. Gemini does not have the tools to do this. It is very useful both for generating summaries of a text and for being able to ask specific questions about the information contained in a file. The only limit is that the document weighs less than 10 MB. We have even surprised ourselves by digitizing books with our mobile phone, and adding character recognition with Adobe Acrobat, so that Claude 3 processes the document and we can create summaries or ask it to explain the most relevant things to us.
Another interesting aspect in which Claude 3 stands out is when it comes to being able to analyze what appears in an image. We have done a test with a photograph and we found the rigor when describing the image great. Although it also seemed to us that he took too much risk when interpreting some element that appeared in the image. Like a text that was not completely visible and that you tried to guess.
We were negatively surprised that Gemini invented things that were not in the photograph. Although none of these artificial intelligences have been able to analyze the exif metadata that appears in each photo, since these are apparently eliminated. We imagine that it is a matter of privacy. Another relevant issue. Gemini cannot analyze images in which people appear, something that both ChatGPT and Claude can do.
A very interesting use of image analysis by these artificial intelligences is to analyze graphics, especially if they are highly complex. This makes things much easier to analyze statistical graphs, for example, and reach conclusions. Another interesting aspect is that all three services can read handwritten text. Although we are forced to issue a warning about Gemini’s notable level of errors when analyzing images. So much so that we even advise against using this feature without verifying the information you provide us.
Many years ago this journalist had the opportunity to see at the Google offices in Zurich the enormous work that the company was doing in AI. So much so that it seemed that the future of artificial intelligence depended on what Google did. And that is partly true, but the Californian company was surprised by the emergence of ChatGPT and had to react quickly by launching Bard, which today has become Gemini. Although Gemini works quite well, and has the quality of being extremely fast in its responses, it still lags behind in terms of precision in the answers it gives us.
But there is one aspect in which Gemini shines: it is the only one of the three AIs that allows us to show updated data thanks to its integration with the Google search engine. This is not something minor. We can ask Gemini things like giving us a summary of the day’s news. Or even specify more and ask him to give us a summary of the war today in Ukraine. To do this, review numerous sources thanks to your search engine. Which makes it very interesting if we are interested in working with updated data. It must be taken into account that Google trained Gemini with data only until May 2022, ChatGPT was trained until April 2023 and Claude until August 2023. So both ChatGPT and Claude are a little out of date with the current situation. And this advantage is what Gemini is exploiting best. That and showing us many more references about where the information it gives us comes from if it comes from the Google search engine.
In fact, using Gemini has a lot to do with the traditional searches we do on Google. Which also makes it easier to use for those who are not used to asking a machine.
It remains to be seen when Gemini is integrated into Android phones. Things could be done like adding an interpreter to translate languages ??that is more advanced than Google Translator. That is probably the future of Gemini and the reason why it beats ChatGPT in speed and Claude emphatically in the speed of its responses. In fact, it is already planned that Google’s most advanced phone, the Pixel 8 pro, could incorporate Gemini Nano. A reduced version of your AI.
ChatGPT unleashed the artificial intelligence fever. AI and chatbots already existed when it emerged just over a year ago. But ChatGPT was much smarter than everything out there. Although it has some limitations. The knowledge base with which it has been trained reaches until April 2023. Claude 3 was trained until August 2023 and Gemini Advance until May 2022. But the latter complements its information with the Google search engine until today. Although ChatGPT 4 still has bugs and can do things as surprising as inventing books that don’t exist, its level of accuracy has improved. And in fact, it is what we think is the most mature AI of the three that we have tested.
In addition, ChatGPT is innovating very decisively. Among other things because it provides developers with tools to use it. That is why ChatGPT has surprised by creating the GPT Store in its paid version. Which is nothing more than a huge compendium of applications that allow you to use services from other companies, developers or Open AI itself using ChatGPT. This allows you to use external services that allow you to generate videos from text or much more surprising things. Such as transcribing entire YouTube videos or asking questions about the content of the video. The GPT Store, which by the way does not sell anything at the moment, in fact makes a big difference with Claude and Gemini.
It is also the only one of the three artificial intelligences that allows you to converse with it out loud. We do not mean that the text can be transcribed using voice recognition. What ChatGPT allows in its app for Android and iOS is to activate a listening mode so that we can talk to her. By the way, if anyone is wondering about Sora, OpenAI’s spectacular text-based video generator using text, it is not available yet and we cannot access it from ChatGPT or anywhere at the moment.
Although things are moving very quickly in the field of artificial intelligence, it is surprising that AI continues to be something primarily designed for use on the computer. Something that is largely due to the fact that if we want to ask complex questions, a computer screen is still much better than a phone screen, but also because we can use a physical keyboard. But that is about to change. ChatGPT has been offering us a lot of useful information for a long time by chatting with it out loud using its app. But we have also seen what AI can do very soon by installing the Gemini version for Android. Although at the moment it is only available in a very restrictive way in the Play Store application store.
This app replaces the Google Assistant, so we can ask it questions and receive the answer with voice using only the command “Ok, Google”. Although it is a little strange that when Gemini goes into action the web browser we use opens showing us the information. We don’t know when Gemini will be available on iOS. But the truth is that we liked what we saw. And this despite the fact that Google, somewhat inexplicably, does not currently encourage its use. To use Gemini in our test we had to download the app from a different app store via the web. If we don’t want to use Gemini that way or we want to use it on an iPhone, we can create a web application and access it from the phone. Something that also works for Claude 3. Just take a look at the instructions from Google and Apple on how to do this on Android and iOS phones and tablets.
Using artificial intelligence on the phone also allows us to be able to interact with what surrounds us. For example, we can take a photo of a series of ingredients that we have in the kitchen and ask the AI ??to check if we are missing something to cook a certain recipe. But we can also get much more detailed information about whatever we’re looking at. Like a building or a monument.