What are the main reasons for the lack of transparency in proprietary LLMs like OpenAI's GPT suite?

Proprietary LLMs like OpenAI's GPT suite lack transparency mainly because their stochastic nature makes outputs difficult to trace deterministically, and because developers withhold training data, methods, and procedures to protect competitive advantage. This lack of disclosure limits researchers' ability to audit or replicate model behavior.

The evidence identifies two interconnected reasons. First, LLMs are stochastic rather than deterministic, meaning they generate responses from statistical representations of language. This makes it hard to relate inputs to outputs in a predictable way or explain why specific content was generated, undermining explainability and therefore transparency. Second, full transparency would require training datasets, training methods, fine-tuning, and inference procedures to be auditable and replicable, but proprietary providers such as OpenAI do not disclose this information because they regard it as a source of competitive advantage. This lack of visibility prevents researchers from understanding model outputs, assessing reliability and potential biases, and tracing the origins of outputs for accountability.