Are We on the Threshold of Artificial General Intelligence (AGI), or Are Crucial Breakthroughs Still Needed?
AI development has accelerated at a blistering pace in recent years, with new records in multimodal learning, increased training speed, and improved mathematical reasoning. But have we already seen the critical breakthroughs required for AGI – or are we still on the way there? In this article, we explore the latest trends, achievements, and challenges in the field, focusing on what it takes to reach true general intelligence. We examine statistics, technical advancements, and safety considerations to give you a deeper understanding of how close we actually are to AGI.
Multimodal Learning: The Key to General Intelligence
One of the most significant trends toward AGI is the development of multimodal AI models. These systems are trained on various types of data, such as text, image, audio, and code, making them far more flexible and powerful than earlier generations of AI.
Multimodal models can integrate and reason across multiple data sources—an ability that closely resembles human intelligence.
Research shows this technology drastically improves AI’s capability to solve complex tasks, for example in medical diagnostics or automated problem-solving.
A clear example is the use of Mixture-of-Experts architectures, which allow for specialized modules within the same model, thereby increasing both efficiency and generalizability. This has enabled AI to achieve as much as 96.7% accuracy on the AIME 2024 mathematics exam—a figure previously unimaginable for machines. AI’s ability to reason with such precision in mathematics is a sign that core AGI components are now in place.
Accelerated AI Development: Speed, Capacity, and Cost
In 2025, a dramatic leap in the speed of AI development occurred: training speeds increased by a staggering 50,000 times compared to previous years. This has not only allowed faster experimentation and iteration, but also enabled the development of more advanced and capable models in record time.
Models that previously required over $100 million in training costs can now be built for just $5.5 million—with comparable performance—thanks to improved algorithms and more efficient software.
This trend toward cost-effective AI development lowers the barrier to entry for more actors and accelerates innovation even further.
For example, a model trained at a fraction of previous costs managed to rediscover state-of-the-art solutions for 75% of 50 open mathematical problems, and improved the solutions in 20% of the cases. This shows that AI can not only learn from existing knowledge but also create new insights—something central to AGI.
Technical Challenges: Data Handling and Security
Despite impressive progress, several critical challenges remain on the road to AGI. One of the biggest is data security within new AI architectures. Traditional Retrieval-Augmented Generation (RAG) systems centralize data, which poses significant security risks:
Centralized data handling can lead to vulnerabilities and leakage of sensitive information.
New security protocols and decentralized solutions are needed to ensure both robustness and privacy.
Several successful methods have recently been introduced. These include deep learning-guided program synthesis and test-time training to reduce reliance on central data stores. These techniques have been crucial in enabling AI models to work securely and efficiently on sensitive tasks in areas like finance and healthcare.
Standardization and International Guidelines
With rapid advances in AI, the need for international standards and guidelines has become increasingly evident. In 2024, there was a shift toward developing global frameworks for “safe, secure, and trustworthy” AI systems. This is critical for managing risks and building trust in the technology.
The lack of expertise is a bottleneck, with only a few leading players driving development—highlighted by the fact that the cost of top-level talent sometimes reaches $100 million.
Standardization helps more stakeholders participate in AI development in a safe and responsible way.
To accelerate AGI development, both technical solutions and shared security frameworks are needed. Companies and researchers are encouraged to engage in international collaborations and adopt recognized standards, such as ISO/IEC AI guidelines.
Summary and Recommendations
So, have the necessary breakthroughs for AGI already been made? There’s no doubt we’re entering a new era of AI with multimodal models, accelerated development, and improved cost-efficiency. At the same time, critical challenges remain in areas like data security and standardization that must be addressed before full AGI is achieved.
Recommendations:
-
Follow developments in multimodal AI models and experiment with integrating diverse data types.
-
Prioritize data security and decentralized solutions in your AI projects.
-
Engage with international standards and collaborations to contribute to safe and fair AI development.
If progress continues at this pace, we may be closer to AGI than many think—but continued innovation and collaboration are required to take the final steps.
This article has been edited with AI support.