From 2020 to 2025, the AI world experienced what Ilya Sutskever, former OpenAI chief scientist, calls the "age of scaling." This era was marked by rapid advancements achieved primarily by increasing compute power, data, and model sizes in a predictable recipe known as scaling laws. The approach fueled major breakthroughs, including the rise of models like GPT-3 and GPT-4, but Sutskever now asserts this era is concluding since scaling alone can no longer guarantee further leaps in AI capabilities.
The grounds for the age of scaling ending lie in practical and theoretical limits. Firstly, the available training data on the internet is nearing exhaustion, and synthetic data augmentation only offers marginal benefits. Secondly, simply enlarging models using more GPUs and data no longer results in substantial improvements in generalization—the models' fundamental ability to learn efficiently and adapt broadly to new situations. This bottleneck of creativity and ideas, rather than raw compute or data, requires new research breakthroughs to overcome.
Looking ahead, the future trajectory of AI is set to return to a research-focused paradigm where novel learning methods, architectures, and scientific insights will drive progress. Sutskever predicts that human-like learning AI with superior generalization will emerge within 5 to 20 years through these advancements. His startup SSI embodies this shift by concentrating exclusively on deep research into reliable generalization and safe superintelligence, avoiding product distractions and competing on innovative ideas rather than scale.
The ultimate goal is to build superintelligent systems that can learn and adapt more like humans, capable of safely understanding and valuing sentient life. Such AI would surpass current benchmarks of raw power and memory, achieving a more profound and reliable mastery over diverse tasks. The age of scaling laid the groundwork, but the age of research will unlock the true potential of AI—ushering in a smarter, safer, and more versatile future.
SCALING BUILT THE FOUNDATION—RESEARCH WILL BUILD THE FUTURE OF AI.