Contribution to the handbook
Supporting · Assessed 2026-10-28
The training details, including the use of LoRA and reported sensitivity and comparison procedures, provide useful technical background for interpreting the study and its tests. They do not by themselves specify a governance system or establish that these methods are suitable for public decisions.
Does this challenge the goal?
In-scope gap
The question concerns model training and comparison methods that can inform implementation and testing within the existing goal; it does not challenge the goal.
The supplied records describe one study’s methods and limited comparisons, not a general basis for selecting a governance model.
Why this category?
- Useful implementation and testing context
- Study-specific methods do not establish general performance or suitability
The packet does not provide the full results of the temperature tests, and the fine-tuning comparison is limited to one model and the study’s selected reference documents.
This is an AI assessment of contribution to the goal, separate from whether the question has an answer.
- What did the researchers report about the GPT-3.5 Turbo training setup?
- How did the researchers try to make model comparisons comparable?
- Did the researchers check whether temperature affected predictions?
- Did a training setting appear to change the reported results?
- Adapting the models
- Temperature tests are reported
- Temperature setting
- Keeping model architecture consistent
- Fine-tuning beat one context method
- Temperature checks
Tests, stop conditions and unresolved work →
Read the exploration goalNew evidence is being assessed. Any answer below is provisional; parts whose evidence changed or was withdrawn are withheld.
Describe the fine-tuning interface and reported architecture status, and distinguish fixed training settings from automatically selected ones.
What we can say so far
The researchers describe building prompts from participants’ details and choices, including both orders of each proposal pair. A GPT-3.5 Turbo comparison describes its fine-tuned model as using LoRA. They characterize GPT-3.5 Turbo’s architecture and original training data as proprietary. They report training at temperature zero for reproducibility and keeping each model’s architecture fixed across training-set sizes. The supplied evidence does not identify the GPT-3.5 Turbo fine-tuning interface or say which training settings, if any, were selected automatically. The study’s general estimate that LoRA adapted about 12% of parameters cannot be assigned specifically to GPT-3.5 Turbo.
What remains open
- The GPT-3.5 Turbo fine-tuning interface or workflow is not specified in the supplied evidence.
- The supplied evidence does not identify which GPT-3.5 Turbo training settings were selected automatically, if any.
Conditions and limits
- These are attributed reports of the researchers’ setup, not independently verified implementation details.
- The prompt-building and approximately 12% LoRA descriptions concern the general model-adaptation method; the parameter figure is not established specifically for GPT-3.5 Turbo.
- The temperature-zero setting and architecture consistency are reported methodological choices, not evidence that all other settings were fixed.
Evidence and related ideas
Supporting evidence, qualifications and context play different roles in this answer.
context
Six fine-tuned models The authors fine-tune six existing language models on the volunteer data: Llama-2 7B, LLaMA-3 8B, ChatGPT 3.5 Turbo, Mistral 7B, Falcon 7B, and Gemma 7B.
Places GPT-3.5 Turbo among the models the authors say they fine-tuned; does not detail its interface or settings.
Large Language Models (LLMs) as Agents for Augmented Democracy
Read full source: Large Language Models (LLMs) as Agents for Augmented Democracy (opens in a new tab) ↗Read source discussion: Large Language Models (LLMs) as Agents for Augmented Democracy →