Based on surveys and interviews, a16z data that analyzed where and how LLM will

Based on surveys and interviews, a16z data that analyzed where and how LLM will be used in the enterprise is summarized.

It is emphasizing the transition from open AI-centered to multi-LLMs that use many open-source and other closed LLMs, including open AI.

"Only six months ago, most businesses were experimenting with either one model (usually OpenAI) or at best two. Today, I have spoken with enterprise leaders and they are all testing multiple models that 1) customize use cases according to performance, scale, and cost, 2) avoid dependency, and 3) can quickly leverage fast-changing field developments, with some using them in production. This third part was particularly important to leaders, as the model leader is dynamic and companies want to integrate both current state-of-the-art and open-source models for best results. We expect more models to spread in the future. Survey data shows that business leaders are testing multiple models, a leading indicator of the models that will be used to push workloads to production. As for production use cases, as expected, OpenAI still holds an overwhelming market share."

"Closed is expected to account for 80% to 90% of the market share in 2023, with OpenAI accounting for most of the share. However, 46% of survey respondents said they would prefer or very much prefer the open-source model in 2024. In interviews, approximately 60% of AI leaders said they were willing to increase the use of open-source or transition when fine-tuned open-source models become almost as close to closed-source performance. After 2024, companies expect a significant shift to open-source, with some clearly targeting 50/50 segments from 80% closed/20% open-source segments in 2023."