Francois Chollet

Future AI in 15 years will not be based on the LLM stack but will necessarily move closer to symbolic learning as its optimal final form

AI will inevitably move to 'patch (2)' in the long term as a more fundamental solution to deep learning's limitations

In the future, AI will not be based on the primitive stack of today, and both training and inference will be incredibly cheap.

Future researchers will investigate model weights from 21st century AI systems to reconstruct cultural information

The era of new model launches as big milestones will eventually come to an end

AI models will be continuously updated with no widely publicized version numbers within less than 2 years

The era of new model launches as big milestones will end, with models being continuously updated instead

Continuous model updates with no version numbers will happen in less than 2 years

The era of new model launches as major milestones will end

Model launches will transition from versioned releases to continuous updates within 2 years

The era of new model launches as big milestones will end within less than 2 years, with models being continuously updated instead

AI model launches as discrete milestone events will end within less than 2 years, replaced by continuous updates without version numbers

Model launches as big milestones will end and models will be continuously updated without publicized version numbers within less than 2 years

AI models will transition to continuous updates without version numbers in less than 2 years

The era of new model launches as big milestones will end, with models being continuously updated without widely publicized version numbers

Model launches as big milestones will end within 2 years, with models being continuously updated without widely publicized version numbers

AI model launches as big milestone events will end within less than 2 years, replaced by continuous updates without publicized version numbers