The past year has been a transformative one for artificial intelligence, characterized by breakthrough innovations, new regulations, and a shift toward practical AI tools that boost productivity. Looking forward to 2025, let us review the major developments in 2024 and discuss future developments.
Part One: 2024 Review
OpenAI is ahead
OpenAI maintains its position at the forefront of artificial intelligence innovation in 2024. This was followed by the launch of the o1 in September, as well as a lighter version, the o1-mini, and the o3 in December.
Perhaps most notably, OpenAI’s Sora project revolutionized video generation. Launched in February and enhanced with Sora Turbo in December, the technology demonstrates unprecedented capabilities in creating realistic video content based on textual descriptions.
Agent AI and AI Assistants
Artificial intelligence in 2024 will not replace jobs, but focus on increasing human productivity through innovative tools and assistants. Google has launched several groundbreaking projects: Astra, an AI assistant for phones and smart glasses, and Mariner, a Chrome extension that enables Gemini to interact with the browser. These tools represent a shift from simple chat interfaces to interactive agents that can understand and manipulate our digital environments.
Anthropic joins this trend by enabling Claude to use the computer like a human – looking at the screen, moving the cursor, and interacting with the interface. This development opens up new possibilities for artificial intelligence to assist with everyday computer tasks.
The developer community is seeing particularly exciting progress. Github’s Copilot and Cursor AI have been well received by the developer community. On the other hand, autonomous coding agents like Devin and Github Workspace are still rough. These tools don’t replace developers; Instead, they work more efficiently by handling daily tasks and suggesting improvements.
Artificial intelligence supervision
2024 saw significant progress in the regulation of artificial intelligence, notably the EU Artificial Intelligence Act. This groundbreaking legislation sets out the first comprehensive framework for the regulation of artificial intelligence, setting out rules for the development and deployment of artificial intelligence systems while balancing innovation with safety and ethical concerns.
Part 2: Looking ahead to 2025
Small, professional model
While pure decoder models such as GPT and Claude dominate the headlines, pure encoder models will still be an important part of artificial intelligence applications in 2025. ModernBERT was released just a week before the end of the year, representing a long overdue major advance in the field. These models are particularly important for retrieval-augmented generation (RAG) settings, where they excel at information retrieval and classification at significantly lower running costs.
In a similar way, current trends in agent systems will favor small, specialized models. These models will be fine-tuned to be as good as cutting-edge models on truly specific tasks, but combined in agent workflows will allow for truly powerful and resource-efficient AI systems.
Multimodal models and test-time calculations
We can expect significant advances in multimodal capabilities, with AI systems becoming more natural at processing and responding to various forms of input (text, speech, images, and video) simultaneously. These improvements will make AI interactions feel more natural and context-aware.
The concept of “compute at test”—giving models more time to think—emerges as a game changer in 2024 and may become even more prominent in 2025. Math and coding benchmarks. They continue to beat humans at very complex and deep reasoning tasks, leaving us wondering just how far general artificial intelligence can go.
Test-time calculations not only show significant results for the cutting-edge model. When used on small edge models, their performance exceeds that of larger models when given more processing time. For example, in the Math-500 benchmark, using a test-time calculation strategy of 256 iterations, Llama 3.2 3B outperforms Llama 3.1 70B.
The evolution of artificial intelligence regulation
Following the EU, we may see more regions implementing AI regulations in 2025.
in conclusion
2024 is a year of significant progress in artificial intelligence. We are seeing many new ways to use artificial intelligence in our daily lives, as well as new ways to improve models, which are taking a big step towards general artificial intelligence. As we enter 2025, the focus seems to be shifting toward more professional and efficient AI systems, improving reasoning capabilities through techniques such as test-time computing, and focusing on AI tools that improve productivity.