The article also predicts that the future of open-source models lies in reducing their size to decrease the cost, complexity, and time required to customize and run them. However, for most applications, the major LLM providers are expected to dominate due to their economies of scale and quality of service. The article concludes that companies trying to compete with these providers would be wasting resources on an unsolvable optimization problem.
Key takeaways:
- OpenAI and other LLM providers have a significant advantage due to their ability to serve models at extreme economies of scale, making it more cost-effective than most organizations running their own open-source LLM deployments.
- These companies collect enormous amounts of data, including user prompts and both explicit and implicit feedback, which can be used for future model training.
- While model quality is a big advantage, the scalability of their infrastructure and the quality of their service are also significant factors in their dominance.
- Open-source models will need to get smaller over time to reduce the cost, complexity, and time required to customize and run them, but for most other needs, major LLM providers like OpenAI will dominate.