A starting point for discussion
The idea
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 participants who identified as left-wing and chose 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, statistical tests found evidence that the models could predict choices, with 65% to 77% accuracy. Overall, the authors conclude that models adapted using LoRA, a method for training a model for a task, predicted individual preferences better than predictions based only on participants’ stated political orientation.
Place in the reasoning
AI assessment · Connected to reasoning
Reports that models outperformed a party-based rule in most—but not all—specified comparisons, with results limited to the studied models, participants, proposals, and comparison methods.
Different prediction checks reveal different limits
The findings concern individual-choice prediction against a particular baseline and include stated exceptions, so they do not establish broad or universal accuracy.
Matching views does not make AI representation accountable
Better prediction than this rule does not itself establish accountable representation in actual civic decisions.
What this means and why it matters
What this means
The reported accuracy favored the models in most comparisons with the party-based rule. For choices between proposals from the same program, where the rule could not make a prediction, the authors also report evidence of predictive ability from statistical tests.
Why it matters
The results suggest that, in these tests, the models captured preferences that this particular rule based on participants’ stated political orientation did not. They do not show that the models would predict choices better in other settings.
Conditions & assumptions
Source coverage: Full readable text of the retrieved 37-page research draft, including references and appendices. Figures are not independently interpreted. The results favor the models in most reported comparisons, but include specific exceptions. The authors also report significant predictive power for choices between proposals from the same candidate’s program. Their overall conclusion is limited to the study’s models, participants, proposals, and comparison methods.