LLM temperature settings significantly affect generated-text length and complexity, with Llama3 producing more complex texts than GPT-3.5¶

Claim · All claims
Evidence · 1 study (2 entries) · causal r1 · q2
Currency · newest study 2025 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q2 i? Higher temperature settings produced longer, more complex generated sentences, with Llama temperature 2 texts averaging 44.4 words versus 25.7 for GPT temperature 0.1. → Keith Cochran 2025 q2 i? One-sided t-tests on Gunning fog readability confirmed higher-temperature generators produced significantly more readable-complex texts for the French question. → Keith Cochran 2025 (2)

Evidence¶

Keith Cochran 2025¶

Keith Cochran, Clayton Cohn, Jean-François Rouet, Peter Hastings. (2025). Comparing Text Augmentation by GPT-3.5 and Llama3 for Evaluating Student Responses. International Journal of Artificial Intelligence in Education. https://doi.org/10.1007/s40593-025-00473-x

q2 · i? · causal · r1

Numerical comparison of average word counts per generator/temperature (Fig. 1): 44.4 words for Llama temperature 2 versus 25.7 for GPT temperature 0.1, against a 15.8-word average student response. Temperature "had a significant effect" on length and complexity.

"The different temperature settings for the models had a significant effect on the length and complexity of the generated sentences."

Keith Cochran 2025 (2)¶

Keith Cochran, Clayton Cohn, Jean-François Rouet, Peter Hastings. (2025). Comparing Text Augmentation by GPT-3.5 and Llama3 for Evaluating Student Responses. International Journal of Artificial Intelligence in Education. https://doi.org/10.1007/s40593-025-00473-x

q2 · i? · causal · r1

Two-sample one-sided t-tests on 1000 randomly sampled texts per generator/temperature for the French question, comparing Gunning fog readability. Differences were "highly significant ( p < .0001)" except GPT 0.9 vs GPT 0.5 (p < .05).

"All the differences were highly significant ( p < .0001) with the exception that texts from GPT 3.5 with a temperature of 0.9 had a mean readability of 13.9 ( SD = 3.8)"

Discussion¶

Related Claims¶

• Higher LLM temperature settings yield the best English augmentation performance, with Llama temperature 2 the top generator overall — related
• The three fine-tuned LLMs differ significantly in response length, readability, and similarity to posts — related
• Prompt-based personalization of an LLM/RAG teaching assistant produces measurable variation in response expression, length, and complexity across personalization conditions — related
• Hyperparameter setting with maximum token 53 and temperature 1 (Set C) yields the most accurate AGE grading performance