Roger Oriol2023 was a great year for AI. Large Language Models were already in the spotlight for both users and businesses. ChatGPT had been just released in late 2022 and was taking the world by storm. Still, 2023 has brought more rapid change in the field than we could have imagined. This last year we got the newest version of OpenAI’s model, GPT-4. We also got a ton of open-source models competing with OpenAI, like LLaMa, Falcon, and Mistral. Google didn’t want to miss the party and stepped up its game unveiling Gemini, the successor to Google’s earlier model, PaLM. Anthropic launched Claude, AI21 Labs launched Jurassic-2 and Amazon announced its Generative AI service, Bedrock, as well as their own LLM, Titan. Some LLMs gained a ton of new functionality by going multimodal, as well as embracing agents, which allow them to have up-to-date information and interact with the world around them.
Also, regulators are beginning to catch up to the rapid evolution of this new technology. The United States of America passed an Executive Order on the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence, and the European Union approved the EU AI Act, the first actual regulation on artificial intelligence.
So, what’s next for 2024? Given the unpredictably fast rise of new AI technology in the last 14 months, any prediction that we make on its evolution in the future will probably be ridiculed by reality. However, an educated guess can be made given the current direction of AI. I invite you to hear about my take on AI in 2024 and of course, take it with a grain of salt.
AGI
As new LLMs begin to approach the trillion parameters mark, new architecture ideas improve on the transformer architecture, and AI’s reasoning abilities continue to improve, I predict that we will begin to hear some Artificial General Intelligence claims. Artificial General Intelligence is the highly anticipated milestone when AI will be able to perform any intellectual task that a human could do. While this kind of artificial intelligence is still very far from being a reality, it seems like it will be inevitable. With current AIs improving exponentially, it's just a matter of time until we begin to hear claims that a top LLM has reached AGI level. While this does not mean that we are there yet, it’s a clear step that we are on the path to that very realistic future.
Open-source models
HuggingFace is a platform that makes it easy to share and contribute to models and datasets. It’s an incredible resource for anyone who wants to start using AI models right away, without having to invest time and money in training their own model from scratch. HuggingFace also provides a ton of resources for working with models and easily fine-tuning them to adapt any model to the user’s needs. In my opinion, these kinds of resources are priceless to both individuals and enterprises that want to get into using AI. And given the fact that every day more and more individuals and businesses introduce AI as a part of their workflows and products, it is likely that this kind of platform will explode further that it already has, providing tons of new open-source models and AI resources.
Multimodality
Some LLMs like ChatGPT and Google’s Gemini have already unveiled multimodal capabilities. Multimodal models are Large Language Models that can understand and interact with a combination of different types of inputs, like text, audio, images, and video. Since LLMs convert text and images to a vector representation to interpret them correctly, it's quite challenging to find a vector representation that works for both text and images, so this is an incredible feat. LLMs having multimodal capabilities broaden the list of tasks they can perform by a large margin. This year, I expect even more models embracing multimodality and the capabilities of AI agents to increase significantly.
Responsible AI
As regulators start to intensify legislation on artificial intelligence, Responsible AI will not just be a choice in developing new AI models, but a must. Responsible AI is an approach to training AI models with ethics and safety in mind. Generative AI models should be trained to be fair and unbiased, as well as to avoid harmful responses and avoid endangering human life, property, or the environment. I expect not only to see an improvement in Responsible AI frameworks and expectations but also to witness the first legal consequences for enterprises not following those new rulings.
AI use in enterprises
As using LLMs becomes easier and more democratized, the barrier of entry for tons of enterprises that are not primarily technologically focused will begin to fall. More and more kinds of companies will begin to use AI to aid in their daily operations. We will see more AI B2B products to address these new necessities. Some companies may just start using AI for the hype of it but will fail to implement it in a meaningful manner for their bottom line. On the other hand, companies that succeed in introducing AI in their daily operations will see an unprecedented boost in productivity. Sadly, this also might mean that we will see some companies lay off an important part of their workforce, as employees are quickly replaced by AI systems.
AI race
The potential of artificial intelligence is no longer a secret. Many countries around the world will not only begin to legislate for AI but also invest heavily in it, as being at the forefront of AI innovation can give them an edge in many areas, including military, scientific, economic, and informational. Even if one specific part of the world advocates for AI ethics and halts AI progress to ensure it not being a threat to humanity, other parts of the world might take this opportunity to put themselves at the front of innovation, which will inevitably make the former’s efforts worthless. We might begin to witness a new AI “space race”, as countries put pressure on each other to never stop investing resources in developing new AI, under the threat of falling behind in the new future that AI is shaping.
Even if my predictions don’t end up becoming a reality, what is certain is that AI will continue to be a major topic in 2024. I’m excited to see how AI will evolve during the following year and what new capabilities it will bring. Certainly, there are many concerns to be had with the rapid evolution of AI, but I’m hopeful that we will be responsible enough to address them before they do become a problem. As a personal note, this year I will be testing how to use AI in many of my daily activities to make them easier or even completely automate them.
