End of Pre-Training Era
At the 2024 Neural Information Processing Systems (NeurIPS) conference in Vancouver, Ilya Sutskever, co-founder of OpenAI, made a thought-provoking statement: the pre-training phase for artificial intelligence models may be nearing its end.
He believes that we are on the cusp of a new era characterized by superintelligent AI.
Sutskever elaborated on this idea by pointing out that advancements in hardware, software, and machine-learning algorithms are outpacing the growth of the data needed to train AI systems.
He compared data to fossil fuels, suggesting that usable data is a limited resource.
Since most data originates from a single source—the internet—he warned that we might have reached a peak in available data, likening it to a depleting resource.
Path to Superintelligent AI
Looking ahead, Sutskever identified several critical developments that he believes will pave the way for superintelligent AI.
Key milestones include the rise of agentic AI, the use of synthetic data, and innovations in inference time computing.
Unlike standard chatbots, agentic AI has the ability to operate independently and make its own decisions.
This capability is increasingly relevant, particularly in the cryptocurrency world.
A notable example is Truth Terminal, a large language model (LLM) that successfully boosted the memecoin Goatseus Maximus (GOAT), leading to a staggering market valuation of $1 billion and catching the eye of everyday investors and venture capitalists alike.
In another exciting advancement, Google’s DeepMind has rolled out Gemini 2.0, a new AI model intended to enhance the development of AI agents.
According to Google, these agents will handle intricate tasks like navigating websites and engaging in logical reasoning, showcasing a significant leap in AI capabilities.
Future of AI and Autonomy
The ongoing progress in developing autonomous AI agents is anticipated to provide a solution to the problem of AI hallucinations.
These hallucinations often occur due to reliance on flawed datasets and outdated LLMs during the training of new models, which can lead to diminishing performance over time.
By focusing on creating more sophisticated AI agents, the industry aims to address these challenges and set the stage for the future of superintelligent AI.
Source: Cointelegraph
