Two very different approaches for LLMs now:
- Two very different approaches for LLMs now:
- 1) The 'Cloud Provider' model: commodity hosting of models.
- Compete on cost.
- The hosting is the point.
- The model is commodity.
- 2) The 'LLM Provider' model: proprietary hosting of models.
- Compete on quality.
- The model is the point.
- The hosting is commodity.
- The latter works best when the model is an order of magnitude better than alternatives.
- But if the proprietary model is only 10% better than commodity alternatives, the benefit goes to competing on cost.
- Especially because even the current generation of models have so much excess quality.
- An embarrassment of riches, if you use them in a way that is resilient to errors.