If we didn't have enough with ChatGPT and Gemini, now Claude emerges, an AI model that promises outperform your competitors and become the reference of the generative artificial intelligence. However, it won't be easy for him. ChatGPT by OpenAI He has the fame and favor of the general public. In addition, it works tirelessly launching new products to stay there. For its part, Google is dedicating time and resources to Gemini to make it the replacement for GPT. In addition, it wants to integrate it and take advantage of it throughout its ecosystem of products, applications and services.
Now there is a third contender. Claude 3, from so far unknown Anthropic, wants to be at the top table with its own AI model. Curiously, he's only been working on it for a year. And, among its promoters, there is also Google. According to his own analysis, his artificial intelligence is better than GPT-4 and Gemini Ultra. Which should make Google and OpenAI increase the speed in the development of their respective AIs. A sector on the rise and in which competition never seems to rest.
Next, we briefly review the technical sheet and curiosities of each of the artificial intelligences in contention. And finally, we will check what the analyzes offered by each of the contenders say: OpenAI, Google and Anthropic. However, and as is usual in any technological field, the users will have the last word of these respective AIs. What is clear is that the competition between them will give a lot to talk about.
What we know about ChatGPT
At this point, it's hard to find someone who doesn't know about ChatGPT. This tool has become a great help for all types of audiences. There are already teachers, high school and university, complaining that their students abuse her. And although it is not perfect, many use it practically daily to carry out any type of task that has to do with process data and generate texts. Programmers, scientists and researchers, content creators, marketing and social media professionals… The list is endless.
ChatGPT is the visible face of GPT, a generative artificial intelligence model. GPT is the acronym for Generative Pre-Trained Transformer. That is, Pre-trained Generative Transformer. The first version of GPT was presented publicly in 2018. Its creators are OpenAI, a company founded in 2015 by Sam Altman, Ilya Sutskever, Greg Brockman, Wojciech Zaremba, Andrej Karpathy and Elon Musk. Of this list, the last three are no longer part of the business organization chart. It was originally a non-profit organization, but in 2019 it changed to a hybrid company. Developments such as DALL-Ewhich generates images through descriptions, and GPT-4which we can use thanks to the chatbot ChatGPT.
GPT is what is known as LLM. Large language model or large language model. From a neural network trained or “fed” with a large amount of text, this AI model learns on its own. One of the usual fields of this type of artificial intelligence is generative artificial intelligence. This AI generates content, whether text, image, audio or video, from the content with which it has been trained and from the requests we make to it through commands.
What we know about Gemini Ultra
Gemini is the Google response to ChatGPT and GPT-4. At the time of writing, the version of Gemini is 1.5. Google has been working on artificial intelligence for years. And it even acquired the company DeepMind, one of the pioneers in this field. From this relationship, artificial intelligence emerged. LaMDA and PaLM. However, OpenAI's strategy of making itself known to the general public with ChatGPT made Google change its strategy. Initially, he launched your own Bard chatbot. But its lack of success caused it to rename its AI product line to Gemini.
Under the name Gemini we find its own chatbot and its own LLM AI model. To facilitate its dissemination, it is available in three versions. From least to most complete. Gemini Nano, Gemini Pro and Gemini Ultra. The latter would be Google's most powerful AI model to date. And according to those responsible, surpasses GPT-4. For the rest, Google is integrating Gemini into its entire ecosystem. Both in the search engine and in your Assistant and in tools such as Gmail, Documents or Google Meet.
What do we know about Claude 3
The third contender is, for now, the most unknown of the three. Anthropic is a company with a few years of life. It was founded in 2021 by siblings Dario and Daniela Amodei. Former OpenAI employees who left the company in 2019 due to their disagreement about partnering with Microsoft. Its star product is Claude 3, a AI model similar to the previous ones. A generative artificial intelligence based on the LLM model. And as a curiosity, it has received billions of dollars from investors of the caliber of Amazon or Google.
Claude 3 is the third version of their AI model. And it is divided into three versions: Claude 3 Haiku, Claude 3 Sonnet and Claude 3 Opus. The strategy is similar to Google's Gemini. Create three versions from least to most complex to adapt to the needs of different audiences and users. The more complexity, the more processing power and the more cost in servers.
Thus, while Haiku It is the lightest version, Sonnet serves for make the chatbot work Claude. At the moment it is free, since it is available in open beta. And, to use it, you need to enter an email address. Or sign in with your Google account. Although if you are looking for more, the paid version will be available in the future, Claude Pro, for 20 US dollars a month. This paid version will use Opusthe most powerful version of Claude.
What the analyzes say about these AIs
Three models of generative AI. All three based on LLM. All three from American companies. Two of them based in San Francisco. While Google resides in Mountain View. Two of them emerged from the OpenAI team. The three AI models have a lot in common, but analyzes and benchmarks should tell us which is better.
Let's start with OpenAI and GPT-4. In its technical pagethose responsible compare GPT-4 with GPT-3.5, present on ChatGPT. Being the first to arrive, they have preferred not to compete directly with others. But that does not exempt the fact that in most of the metrics provided, for example, PaLM, Google's AI model. EITHER Calls, from Meta (formerly Facebook). For example. In a test of 14,000 multiple-choice problems on 57 different topics, “in 24 of the 26 languages tested, GPT-4 outperforms GPT-3.5 in English and other LLMs such as Chinchilla or PaLM.”
To demonstrate the Gemini Ultra superiorityGoogle compare directly your AI model with GPT-4. Your main contender. In text-related tests (general, reasoning, math, and code), seven of the tests are outperformed by Gemini Ultra 1.0 with respect to GPT-4. In it image, video and audio processingall the evidence seems to favor Google over OpenAI. Here You can read the complete technical document with all the analyzes and benchmarks carried out.
What the Claude 3 benchmarks say
Finally, Anthropic joins the battle with your own results table. While GPT-4 started as the favorite in its own analyses, and was seen outperformed in almost all analysis categories by Google's Gemini Ultra, the third rival appears to have surpassed its competitors. According to data provided by his own evidence. The Anthropic table includes its own AI models, that is, the three versions of Claude 3. As well as GPT-4, GPT-3.5 (which is the one used by the free version of ChatGPT and Microsoft CoPilot), Gemini 1.0 Ultra and Gemini 1.0 Pro. These last two are the most complete versions of Google's Gemini.
Based on the analyzes carried out in which the three contenders appear, ChatGPT, Claude and Gemini, the winner is the winner in all the tests Claude 3 Opus. That is, the most powerful version of this AI model. For its part, Gemini 1.0 Ultra It doesn't come out well at all. Although it would surpass its smaller rivals, Claude 3 Sonnet and Haiku. But it remains to be seen what the result would be if we had included in the tests Gemini 1.5 Pro. OpenAI's GPT-4 would be in third position in most tests, with some exceptions. Outperforms Gemini on multiple choice tests and common knowledge.
As I said at the beginning of this article, the final answer will be the users of these AI models. They will know how to tell which AI is the best and which offers the best results. On the other hand, we will be attentive to future testing and analysis by OpenAI and Google. It is inevitable that they will include Claude 3 in their results to compare their improved versions of GPT or ChatGPT and Gemini.