Not necessarily 'exponential' (more superlinear) in capabilities (yet) but rather in parameters/training data/compute/costs, which may sometimes be confused for the other.

[0]: https://ourworldindata.org/grapher/exponential-growth-of-par...

[1]: https://ourworldindata.org/grapher/exponential-growth-of-dat...

[2]: https://epoch.ai/blog/trends-in-training-dataset-sizes

[3]: https://ourworldindata.org/grapher/exponential-growth-of-com...

[4]: https://blog.tebs-lab.com/p/not-exponential-growth