The next generation of artificial intelligence will arrive by the end of 2024! GPT-4, Amazon Olympus, Grok 2, Google Gemini 2… discover the new AI models and their incredible power in the preview!
In 2017, Google researchers published their study “ Attention is all you need » introducing a deep learning architecture called Transformer. It was the beginning of a revolution in the field of artificial intelligence.
It’s on this the architecture on which models like GPT are basedthe first version of which he launched OpenAI in June 2018
Since then, technology has developed very quickly with the appearance of numerous Transformer models and significant increases in power between each generation.
However, if you are already excited from GPT-4 Turbo and other state-of-the-art artificial intelligence, know that you haven’t seen anything yet. The next generation of models is coming, i artificial intelligence that will be published by the end of 2024 it will be much wider, more powerful and ever closer to human intelligence…
Through this file, you will discover the most anticipated LLM (large language models) and a glimpse of the colossal power it will deliver!
Amazonian Olympus: a giant from Cloud enters the AI race
Artificial intelligence is the new front in the GAFAM war, and Amazon does not intend to stand aside.
As of November 2023, we know that the company of Jeff Bezos formed a team of experts responsible for create your own LLM: Olympus.
WITH over 2 trillion parametersit is directly placed among the largest existing models in the range of Claude 3 Opus Anthropic. In comparison, GPT-4 has 1.75 trillion parameters.
l’the team is led by Rohit Prasad, previously in charge of the voice assistant Alexa. Its researchers will work with Amazon’s scientific team to create an AI that can compete with the best models on the market.
Even if Amazon invests heavily in the startup Anthropic and his AI Claudeas in other companies, creating your own model is strategically important. This can especially strengthen the appeal of the AWS cloud.
This AI will be they are mostly sold to companies, but it can also be used to improve the performance of the Alexa and Echo range of devices. Its release date is expected in August 2024.
xAI Grok 2: Elon Musk’s AI that rocks GAFAM
While GAFAM would dream of sharing AI cakes, Elon Musk is the prey and don’t plan on letting them sit around the table by themselves.
Annoyed by the awakened atmosphere of Silicon Valleyhe started his own company xAI and his LLM Grok the first version of which was launched in March 2024.
His goal? Create artificial intelligence focused on the search for truthinstead of serving a political and ideological agenda like Google a Gemini who refuses to represent a white man in a positive sense.
So far, Grok lags behind the competition with only 314 billion parameters. However, Musk announced the Grok 2 and promises a great increase in its capabilities.
While The Grok 1.5 is expected to launch by mid-2024with better comprehension and reasoning skills, Grok 2 could beat it the best AI on all current benchmarks.
Her the training has already started, but we don’t yet know the colossal size this LLM will reach. During an interview with Norwegian investor Nicolai Tangen, Musk said he would it needs 20,000 GPUs Nvidia H100 train this artificial intelligence.
Its release date is scheduled for 2025 and Elon is already expecting Grok 3 which in turn could require more than 100,000 Nvidia H100…
OpenAI GPT-5: the return of the godfather of industry
With GPT-3 in late 2022, then GPT-4 in March 2023, OpenAI is the company that brought generative AI and LLM in the spotlight.
Even today it is considered an American company reference even if some AI rival is GPT-4like Claude 3 Opus or Google Gemini Ultra Pro.
However, the entire industry has its eyes on GPT-5 : the next version of the model, which could offer a real leap forward and risks ” drive in competition » according to the CEO Sam Altman.
So far, few details are known about this new artificial intelligence, because OpenAI insisted on secrecy. We know that she likely to be multimodale, capable of processing text as well as images, sound or even video.
It should also be brought huge performance gain for thinking, coding, knowledge or conversation.
Her training started in 2023, and the number of parameters should be significantly above one and a half trillion GPT-4. It is also possible that the new architecture allows for a significant increase in performance even at the same size.
Instead of a simple chatbot as in previous versions, GPT-5 can assign tasks to submodels and connect different services to perform actions.
Instead of simply asking the AI questions, you can, for example, ask it make phone calls, book tickets plot or create spreadsheet sheets from data.
Please note thatThese are just rumors at the moment. and logical assumptions. GPT-5 is planned for 2024, and Sam Altman said that this AI will surprise everyone…
In April 2024, Meta was launched Llama 3: The third version of its AI LLM open source. However, initially, she launched only two first models this new family.
THE the first is Llama 3 8B with 8 billion parametersAND Llama 3 70B with 70 billion parameters. Both outperform the competition in their categories in terms of thinking, knowledge, skill acquisition, mathematics or code generation.
So, Llama 3 8B proved to be more effective than Mistral 7B or Google Gemma 7B, doc Llama 3 70B competes with Gemini 1.5 Pro, GPT 3.5 or even Claude Sonnet. However, they remain limited by their number of parameters.
In the future, much more powerful Llama 3 models will launch Meta. The company is currently developing a version with more than 400 billion parameters.
She will be able to talk in multiple languagesfor processing more data, etcunderstand various modalities like pictures and videos…
Google Gemini 2: the second version that will crush the competition?
Originally launched in December 2023. Gemini is a family of AI models created by Google. Includes Gemini Nano, Gemini Pro and Gemini Ultra.
Their strong point? Multimodality, which allows them to they process text, but also images, audiovideos or code.
In February 2024. Gemini 1.5 and its Pro version were launched is among the most successful LLMs of the moment. This model particularly shines with its understanding of context, especially thanks to new MoE (Mixture-of-Experts) architecture..
Her the context window reaches 1 million tokens, allowing it to process 1 hour of video, 11 hours of audio, 30,000 lines of code, and more than 700,000 words. In comparison, GPT-4 Turbo manages with 128,000 tokens.
However, Google doesn’t intend to stop there. Training for Gemini 2 has already begun, and its launch is scheduled for November 2024.
THE performance should be significantly increasedacross all modalities, and this AI might even be capable of advanced reasoning.
With this new artificial intelligence, Google may catch up on the competition of other GAFAM and establish itself as the new champion in the industry. However, you’ll have to wait a few months to find out for sure!
YOUR now i know what to expect for the second half of 2024 and the beginning of 2025. A new wave of artificial intelligence will sweep the world, and The AI revolution has only just begun !
and you What is your favorite artificial intelligence at the moment and are you most looking forward to? Express yourself in the comments!
- Share the article:
Our blog is powered by readers. When you shop through links on our site, we may earn an affiliate commission.