Google started the race to develop artificial intelligence (AI) chatbots at a slower pace than rival OpenAI, but the gap is closing.
The tech giant released an upgraded version of its platform, Bard, on May 10 in an effort to give users features that ChatGPT-4 has, and even better ones.
The differences between the two systems are still perceptible, but experts believe that the race is catching up.
“At the moment ChatGPT-4 has a clear advantage. Bard is coming as a challenger, not as a leader”, considers Ricardo Carreón, director of technology at Hexabit, an AI firm based in Texas, USA.
“But it’s going to get close and it’s going to be real competition between the platforms. The issue will be how the user base evolves in both. In the end, what is at stake is who is going to have more users,” Carreón told BBC Mundo.
One of Bard’s main limitations remains that it only accepts instructions in English, Japanese, and Korean. To date, the promised adaptation to more than 40 languages with which the new version was presented has not arrived.
“As we continue to responsibly develop Bard, we are gradually expanding access to more countries and regions in English. We will continue to implement it over time,” a Google source told BBC Mundo, without specifying dates.
On the contrary, one advantage it has over ChatGPT-4 is that the latter is a paid version (US$20), while Google’s chatbot is free to use.
Other than that, how close or far are the new Bard features compared to ChatGPT-4?
1. The images in Bard
Until a few months ago, interaction with Google and OpenAI chatbots was limited to text.
In an era in which the Internet favors the visual, this apparently did not fit. But OpenAI broke the barrier by introducing ChatGPT-4 in March with the integration of images to its chatbot.
Bard has matched that feature in its new version, something that picks up the pace of the competition.
The most basic function is to receive instructions based on an image. For example, you can ask where that piece of art someone shared on Instagram is located without mentioning the source.
On the other hand, there is the offer of answers in images. With this, for example, you could ask him to present you with photos of the most visited paintings in the Louvre Museum in Paris and you will get the answer with text and photos in seconds.
But the innovation of AI goes further: “It is not just looking for a photo, it is reading what is in the photo and based on that, being able to execute a certain action,” explains Carreón.
“If I have eggs, tomato, onion on the table and I ask him ‘what can I do?’, he can suggest some Mexican-style eggs and show me the recipe with those ingredients. Then it already recognizes what is in the photo and goes beyond looking for similar images. It’s understanding what’s in them.”
This ability is called “multimodal function”: instead of giving the chatbot instructions only in text, it is also capable of understanding images.
Another feature that is not yet available, but has been announced, is Adobe’s AI image generation. When it is launched, you could ask it to create the image with the combination of elements that you indicate.
A photo of your pet wearing a birthday hat? It’s something you could ask for.
Google says that in this regard they are still in the experimentation phase. “We want to implement it gradually and responsibly,” said the source consulted.
It would be a huge advance over ChatGPT-4, which has not revealed plans for a similar feature.
2. Integration to other services
Having the information that AI chatbots offer is a first step, but what next?
In the new version of Bard, Google now offers a direct “output” of that information to its cloud services.
You can directly bring the content to your Gmail if you ask him to write a vacation request for your boss. Or to the Docs service if you asked him to compose a song for your dad.
The OpenAI chatbot only offers a button to copy the content to the clipboard.
“At the end of the day, the Google world is fighting the Microsoft world with ChatGPT-4. He has also announced it as the co-pilot in email, in Office, for programmers on GitHub, it will be your co-pilot even in Windows”, warns Carreón.
“So in certain Windows functions you will be able to resort to an artificial intelligence assistant to do certain tasks within the operating system,” he adds.
What is really expected of the AI industry is not only to create chatbots, but also to adapt this enormous capacity for information processing and learning to other applications for daily or specialized use.
“With a restaurant reservation application, by integrating intelligence you can already tell it in the chat to recommend a place that is vegan, that is open at night until 00:00. He gives you the list and you tell him where you want to book”, says Carreón.
3. Sample of information sources
Both Google and some experts have highlighted the improved capabilities of mathematical reasoning or coding in more than 20 programming languages that the new version of Bard offers.
But this is only perceived by professionals in those areas.
However, one additional feature of Bard that has been highlighted for regular user use is the font listing which is now done by the Google chatbot.
The AI has raised concerns about the “hallucinations” that these chatbots come to present. This is the name given to the erroneous, inaccurate or deficient information that they produce from their huge data banks and their own ability to produce it based on learning.
Bard now displays the multiple sources from which it pulled information, something ChatGPT does not. This helps build confidence in the results, which can be verifiable.
For Carreón, this is valuable, but he warns that you have to be cautious and have a critical judgment about what the chatbot presents.
“AI accesses information from its data pool. But it is also true that it generates information, because how the model learns. There is going to be some information that is not coming from a particular place,” he explains.
“If they have already learned that there is a good solution for a problem, they will not necessarily give you links. But since there is still a certain error in these programs, it is very important that all the people who use the AI verify the final result, because sometimes it has errors”.
For the Hexabit expert, users should be clear that AI is not a mere information seeker.
The goal of the developers has been to create a tool with the ability to imitate human learning and reasoning.
“As Bill Gates says, search as we know it today is likely to disappear. And that has huge implications, not just for Google, but for all content publishers that use Google as their main traffic platform,” he says.
“I think many people don’t realize yet that this is going to change everything, but in a brutal way. And it is already happening, there are already many things that are changing ”.
Remember that you can receive notifications from BBC Mundo. Download the new version of our app and activate them so you don’t miss out on our best content.
- Do you already know our YouTube channel? Subscribe!
See original article on BBC