Contribution to the handbook
Supporting · Assessed 2026-10-28
The comparison and its exceptions clarify what the study’s performance results do—and do not—show. This is useful evidence background, but does not alone establish a governance mechanism or system readiness.
Does this challenge the goal?
None identified
The question compares study benchmarks and does not dispute the exploration’s goal or scope.
The results concern the study’s participants, proposals, models, and comparison methods, not other civic settings.
Why this category?
- meaningful comparison
- performance limits
- exceptions
The authors’ interpretation that models capture preference details beyond political alignment is not proof of what information the models used.
This is an AI assessment of contribution to the goal, separate from whether the question has an answer.
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 Any answer below is provisional; parts whose evidence changed or was withdrawn are withheld.
Compare model accuracy with the party-based rule, including material exceptions to the overall pattern.
What we can say so far
In the reported study, the models were more accurate than the party-based rule in 21 of 24 comparisons. The exceptions were liberals choosing proposals from Bolsonaro’s program with Mistral, Falcon, and Gemma. For left-identified participants choosing proposals from Lula’s program, LLaMA-2 7B was 15 percentage points more accurate than the rule; all six models achieved at least 80% accuracy, versus 71% for the rule.
What remains open
- No overall numerical accuracy margin across the 24 comparisons is supplied.
Conditions and limits
- These are reported results for the study’s models, participants, proposals, and comparison methods; performance elsewhere is not established.
Evidence and related ideas
Supporting evidence, qualifications and context play different roles in this answer.
supports
Models outperformed a simple rule The models were more accurate than the party-based rule in 21 of 24 comparisons. The exceptions were cases where liberals chose proposals from Bolsonaro’s program for Mistral, Falcon, and Gemma. For left-identified participants choosing proposals from Lula’s program, LLaMA-2 7B was 15 percentage points more accurate than the rule; all six models reached at least 80% accuracy, compared with 71% for the rule. For choices between proposals from the same candidate’s program, the models had significant predictive power, with 65% to 77% accuracy. Overall, the authors conclude that LoRA fine-tuned models predicted individual preferences better than predictions based only on participants’ self-reported political orientation.
Supplies the 21-of-24 result, three exceptions, and quantified Lula-program subgroup comparison.
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 →
supports
Models beat a simple party rule The models also predicted group preferences more accurately than a rule that assumes people always choose the policies of their preferred party or politician.
Compare model predictions with predictions based only on participants’ political orientation.
What we can say so far
In the study, the models generally predicted individual choices more accurately than a rule based only on participants’ self-reported political orientation. They outperformed that rule in 21 of 24 reported comparisons; the exceptions involved liberals choosing proposals from Bolsonaro’s program for Mistral, Falcon, and Gemma. For left-identified participants choosing proposals from Lula’s program, LLaMA-2 7B was 15 percentage points more accurate than the rule, and all six models reached at least 80% accuracy versus 71% for the rule.
What remains open
- No overall numerical accuracy margin over the orientation-based rule is supplied.
Conditions and limits
- The comparison concerns the study’s models, participants, proposals, and specific rule based on self-reported political orientation.
- The numerical accuracy figures describe left-identified participants choosing proposals from Lula’s program, not an overall accuracy margin.
- The supplied accounts derive from the same preprint and are not independent corroboration.
Evidence and related ideas
Supporting evidence, qualifications and context play different roles in this answer.
supports
Models outperformed a simple rule The models were more accurate than the party-based rule in 21 of 24 comparisons. The exceptions were cases where liberals chose proposals from Bolsonaro’s program for Mistral, Falcon, and Gemma. For left-identified participants choosing proposals from Lula’s program, LLaMA-2 7B was 15 percentage points more accurate than the rule; all six models reached at least 80% accuracy, compared with 71% for the rule. For choices between proposals from the same candidate’s program, the models had significant predictive power, with 65% to 77% accuracy. Overall, the authors conclude that LoRA fine-tuned models predicted individual preferences better than predictions based only on participants’ self-reported political orientation.
Reports the 21-of-24 comparison, exceptions, and quantified Lula-program subgroup comparison.
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 →