OpenAI co-founder Ilya Sutskever recently stated that AI researchers must find new ways to expand machine intelligence in order to overcome existing limitations.
Sutskever delivered a speech at the 2024 Neural Information Processing Systems (NeurIPS) conference in Vancouver, Canada, where he argued that the era of AI pre-training is coming to an end and predicted the rise of superintelligent AI.
Sutskever believes that the increase in computing power through better hardware, software, and machine learning algorithms has already outpaced the total amount of data available for training AI models. He likened data to fossil fuels, which will eventually run out. Sutskever stated:
“Data will not grow because we only have one internet. You could even say that data is the fossil fuel of AI. It was created in some way, and now we are using it, and we have reached a peak in data availability—there won’t be more data in the future; we must work with what we have.”
This OpenAI co-founder predicts that agent AI, synthetic data, and reasoning-time computation will be the next evolutionary directions for AI, ultimately leading to the emergence of superintelligent AI.
Agent AI May Disrupt Existing Models
Agent AI capabilities surpass current chatbot models, enabling them to make decisions without human intervention. With the rise of large language models (LLMs) like AI meme coins and Truth Terminal, agent AI has become a hot topic in the cryptocurrency field.
Truth Terminal quickly gained popularity after launching a meme coin called Goatseus Maximus (GOAT). This meme coin eventually reached a market value of $1 billion, attracting the attention of retail investors and venture capitalists.
Google's DeepMind AI lab has released Gemini 2.0—an AI model that will power agent AI.
According to Google, agents built using the Gemini 2.0 framework will be able to assist with complex tasks, such as coordination and logical reasoning between websites.
The advancements in AI agents that can act and reason independently will lay the groundwork for AI to overcome data hallucinations.
The emergence of AI hallucinations is due to incorrect datasets and the increasing reliance of AI pre-training on using old LLMs to train new LLMs, which can degrade performance over time.
Data Bottlenecks and the Future of AI
Sutskever's views highlight the significant challenges facing AI development: as AI models continue to scale, the demand for data grows. However, the reality is that the growth rate of available data lags far behind the increasing demand for data by models. This forces researchers to explore new methods to overcome data bottlenecks.
Agent AI, synthetic data, and reasoning-time computation may become new directions for the future development of AI. These technologies are expected to help AI models reduce their reliance on vast amounts of data and enhance their reasoning and decision-making capabilities. The emergence of superintelligent AI signifies that AI technology may usher in a new era, potentially transforming our current ways of living and working.
However, the rise of superintelligent AI also raises concerns about AI ethics and safety. How to enjoy the conveniences brought by AI technology while ensuring its controllability and safety will be a question we need to seriously consider.
