The pursuit of artificial general intelligence (AGI) has become an expensive endeavor for many developers. A new approach is emerging that focuses on building lightweight and ultra-cheap AI models, making it more affordable to use AI apps.
By Kai-Fu Lee
The pursuit of artificial general intelligence (AGI) has become an expensive and unaffordable endeavor for many application developers.
The Current State of LLMs
The current state of large language models (LLMs) is characterized by a lopsided ecosystem that is bottom-heavy and top-light. The LLMs trained by the largest GPU farms are usually very expensive for inference, making it difficult for application developers to proliferate killer apps.
The Cost of Inference
The cost of inference has fallen by a factor of 10 per year, pushed down by new AI algorithms, inference technologies, and better chips at lower prices. However, the current cost of using OpenAI’s top-of-the-line models is still very high, with a price tag of around $1 per query.
A New Approach
A new approach is emerging that focuses on building models that are almost as good as the top LLMs but are lightweight and ultra-cheap. This approach will enable application developers to build AI apps that are not only affordable but also perform well.
Vertical and Deep Integration
Vertical and deep integration can optimize inference, model, and application development holistically. For instance, Rhymes.ai trained a model almost as good as the best from OpenAI for $3 million compared to the more than $100 million that Sam Altman said it cost.
A New Law for AI Inference
A new law for AI inference is just around the corner, with the cost of inference falling by a factor of 10 per year. This has been pushed down by new AI algorithms, inference technologies, and better chips at lower prices.
Achieving Equilibrium
The ecosystem must work together to get over the cost hurdle and adjust the formula, achieving equilibrium to make AI really work for our society. Vertical and deep integration that optimized inference, model, and application development holistically was key to achieving this result.
Key Points
-
The pursuit of AGI has become an expensive and unaffordable endeavor for many application developers.
-
A new approach is emerging that focuses on building lightweight and ultra-cheap models.
-
The cost of inference has fallen by a factor of 10 per year, making it more affordable to use AI apps.
-
Vertical and deep integration can optimize inference, model, and application development holistically.
-
The path to AGI requires a focus on building powerful, affordable, and scalable models.
