We’ve tried to predict what’s next in AI for the past few years. It’s a risky move due to how quickly this field changes. But hey, we’ve been on a roll, so we’re doing it again.
How did we do last time? Our four hot trends in 2024 feature customized chatbots. These tools are intelligent assistants powered by advanced multimodal language models. (We didn’t know it back then, but we were talking about today’s AI agents—the hottest trend in the field.) We also predicted generative video would take off. Check this out! OpenAI released Sora. Google DeepMind launched Veo a week later in December. We also noted robots that can perform many different tasks. Large language models are helping to move the robotics field ahead quickly.)
Speaking of gaming, AI tools appear even in unlikely places, like online platforms such as Betlabel Casino, where personalization, smart predictions, and responsive interfaces are becoming part of the experience.
So, what’s coming in 2025? Let’s skip the obvious—yes, agents and leaner language models will keep evolving. But here are five less expected AI trends we think will make waves in 2025.
Large Language Models that “reason.”
The buzz was justified. In September, OpenAI introduced O1. This marked a shift in how large language models function. Just two months later, the company took it further by releasing o3. This model could change the technology for good.
In December, Google DeepMind introduced Mariner, a prototype web-browsing agent. During a live demo for MIT Technology Review, product manager Megha Goel asked it to find a Christmas cookie recipe that matched a photo. She wanted one that matched a photo she provided. Mariner found a recipe online. Then, he added the ingredients to Goel’s grocery basket.
It was a remarkable moment. The agent didn’t hit a wall. Instead, they broke the task into steps. Rather than freeze up, the agent broke the problem down and picked a specific action to move forward. Clicking a browser’s Back button might sound simple, but it’s surprisingly complex for an AI. It worked! Mariner checked the recipe, noted the flour type, and continued filling Goel’s basket.
Of course, OpenAI and Google aren’t alone. Many other players are working on their advanced language models. They use similar reasoning methods to improve their skills, which helps with tasks like cooking and coding. Reasoning is the next big frontier. In 2025, it will be one of the most talked-about AI capabilities.
It’s Boom Time for AI in Science
One of the most promising roles of AI is its ability to speed up scientific breakthroughs. In October, the Nobel Prize in Chemistry honored the makers of AI tools such as AlphaFold. Their work has made a big difference in molecular biology. Demis Hassabis and John M. Jumper from Google DeepMind won for AlphaFold, a tool that helps solve protein folding. David Baker was also honored for his work on tools that aid in designing new proteins. This shows how AI can make a big difference in science.
Look for this trend to continue next year. We will probably see more AI models and datasets for scientific research in the future. Proteins are ideal for AI because they provide great data sets for AI model training.
The race is on to discover what AI can help unlock next. One potential area is materials science. Companies like Meta have released large datasets and models. These aim to speed up discoveries in material science. In December, Hugging Face joined forces with Entalpic. They launched LeMaterial, an open-source project. Its goal is to make research on materials easier and faster. This open-source project aims to make research on materials easier and quicker. Their first project is a dataset. This platform organizes and improves primary materials datasets. This makes it easier for AI systems to learn from them.
AI developers are increasingly positioning their tools as valuable resources for scientists. AI isn’t ready to replace researchers yet, but it’s becoming more useful in labs.
AI Companies Get Cozier with National Security
Military partnerships, which focus on border surveillance and intelligence work, are also becoming a significant source of revenue for AI companies.
The U.S. military is investing heavily in AI and using it in defense operations. Projects like the $1 billion Replicator drone and the AI Rapid Capabilities Cell demonstrate this.
By 2025, defense tech firms such as Palantir and Anduril will continue to expand rapidly. These efforts give companies access to unique military data, which helps them enhance and train their AI systems. Top tech firms such as OpenAI, Microsoft, Google, and Amazon are teaming up with the Pentagon.
OpenAI has teamed up with Anduril to create AI that can take down drones. This marks a significant change from its previous choice to stay out of military projects. Their reasoning? Helping democracies lead in AI development.
As AI development costs go up, more companies might pursue defense contracts. This could happen even if it conflicts with their values. In 2025, expect more debate and deals in the AI and defense world.
