Dokumentationerreichbar
OpenAI, Optimizing LLM Accuracy
developers.openai.com (externe Seite)
Open
geprüft 10.11.2026
Worauf sich diese Seite beruft, wörtlich, abgerufen am 21.10.2026:
these are all levers that solve different things, and to optimize in the right direction you need to pull the right lever
bestätigt 10.11.2026the model lacks contextual knowledge because it wasn’t in its training set
bestätigt 10.11.2026This axis maximizes response accuracy
bestätigt 10.11.2026the model is producing inconsistent results with incorrect formatting
bestätigt 10.11.2026This axis maximizes consistency of behavior
bestätigt 10.11.2026Prompt engineering is typically the best place to start
bestätigt 10.11.2026with even the 3.5 model outperforming few-shot gpt-4 by 8 points
bestätigt 10.11.2026RAG actually decreased accuracy, dropping four points from our GPT-4 fine-tuned model to 83.
bestätigt 10.11.2026many of our largest customer deployments at Open
bestätigt 10.11.2026AI were done using only prompt engineering and RAG Start with 50+ examples
bestätigt 10.11.2026fine-tune the model with RAG examples in it so it isn’t learning how to use the context zero-shot
bestätigt 10.11.2026With fine-tuning you need to rerun the fine-tuning process and manage your training and validation sets when you do additional tuning.
bestätigt 10.11.2026squeeze as much accuracy out of basic methods as you can before reaching for more complex RAG or fine-tuning
bestätigt 10.11.2026If we have a set of 20+ questions and answers
bestätigt 10.11.2026We assume this happens in 5% of cases.
bestätigt 10.11.2026adds 1000 examples from outside of the training and validation sets which are embedded and placed in a vector database
bestätigt 10.11.2026