Hybrid AI in 2024
In 2024, artificial intelligence will be dominated by "hybrid AI" where large "foundation models" like GPT-4 are used as the brain/orchestrator, while smaller specialized AI models focus on specific tasks and tie into the foundation models. Building the large foundation models will only be possible for the richest tech companies like OpenAI, while most enterprises will build smaller specialized models. There will also be a rise of third-party applications built on top of these foundation models.
Summary
- In 2024, artificial intelligence will be focused on "hybrid AI" systems combining large foundation models like GPT-4 with smaller specialized AI models for specific tasks that connect to the foundation models.
- Building huge foundation models will only be possible for the very richest tech companies like OpenAI - most enterprises will build smaller specialized models, either on-premises or in the cloud, trained on their own data.
- There will be a need to coordinate security across different parties building these hybrid AI systems.
- 2024 will also see the rise of many third-party applications built on top of the large foundation models, akin to applications built on top of an operating system.
- It will be important to establish principles and rules around issues like bias, discrimination, safety, privacy etc. to allow AI progress while managing risks.
- The CIO believes fostering collaboration between players, encouraging open sourcing of models where appropriate, and developing principle-based rules will allow the US to continue leading in AI development.