I stumbled upon this fascinating article by Stuart Mills looking at the challenges that further development and operations of AI models face.
The costs of model development and operation are increasing. Efficiencies in development and operation are challenging but may be addressed in the future. However, model quality remains a significant challenge that is more difficult to solve.
Data is running out. Solutions such as synthetic data also have their limitations.
There is also a severe challenge around chips. There is a supply shortage in the context of geopolitical tensions between China, the US, and the EU. Also, the environmental costs of running large AI models are significant.
![](https://i0.wp.com/niekdegreef.nl/wp-content/uploads/2024/06/IMG_3462-September-20-2022-3024-x-4032-kopie.jpg?resize=600%2C800&ssl=1)
The costs of model development and operation are increasing. Efficiencies in development and operation are challenging but may be addressed in the future. However, model quality remains a significant challenge that is more difficult to solve.
Data is running out. Solutions such as synthetic data also have their limitations.
There is also a severe challenge around chips. There is a supply shortage in the context of geopolitical tensions between China, the US, and the EU. Also, the environmental costs of running large AI models are significant.
Two revenue models may emerge in the AI industry. Each with their own take on the cost aspects highlighted above. The first is the ‘foundation model as a platform’ (OpenAI, Microsoft, Google), which demands increasing generality and functionality of foundation models.
The second is the ‘bespoke model’ (IBM), which focuses on developing specific models for corporate clients.
Government action can support and undermine the AI industry. Investment in semiconductor manufacturing in the US and China may increase the supply of chips, and strategic passivity from governments around regulations such as copyrights is suitable for the industry. Government interventions should regulate the AI industry in areas related to the socially and environmentally damaging effects of data centers, copyright infringement, exploitation of laborers, discriminatory practices, and market competition.