Temperature (AI parameter)
Temperature is a hyperparameter that controls how much randomness goes into an LLM's responses. A low value (close to 0) produces deterministic, predictable answers. A high value (close to 1 or above) generates more varied but less reliable answers. For legal use, a low temperature (0.1-0.3) is recommended to ensure accuracy and fidelity to the sources.
Temperature is one of the most important parameters to understand if you want to use AI responsibly in legal work. This setting, whose values typically range from 0 to 2, controls how much randomness goes into the model's choice of the next token (a word or part of a word). At temperature 0, the model consistently picks the most probable token: responses are deterministic and reproducible. At a high temperature, the model explores less likely alternatives, which adds creativity but also unpredictability.
For legal professionals, the temperature setting is critical. A low temperature (0.1 to 0.3) is recommended for any legal task that requires accuracy and fidelity to the sources: contract analysis, case law research, summarising decisions. At these levels, the model stays as close as possible to factual information and minimises the risk of hallucinations.
A higher temperature can nonetheless be useful in certain legal contexts: brainstorming arguments, exploring litigation strategies, or drafting first versions where creativity is welcome. The key is to deliberately choose the temperature level based on the task, and to always review the AI's output regardless of the temperature used.
Related terms
An LLM (Large Language Model) is an artificial intelligence model trained on massive volumes of text to learn the statistical stru…
AI hallucinations refer to the generation of false, fabricated or inconsistent information presented with apparent certainty by a …
Prompt engineering is the art of crafting precise, structured instructions to get optimal results from a generative AI. It is not …
GPT (Generative Pre-trained Transformer) is a family of AI models developed by OpenAI, built on the Transformer architecture intro…
Claude is an AI model developed by Anthropic, a company founded in 2021 by former OpenAI researchers. It stands out for its "Const…