Skip to content
OpenTrain AIFor AI Companies

HFEPX · Eval paper review

When Self-Consistency Backfires: Majority Vote Hurts the Majority of Hard Science Problems for Small LLMs

Utkarsh Bahuguna

Published

Aug 11, 2026

Citations

0

Trust level

High

Usefulness score

75/100 (High)

Extraction confidence

90% (High)

Derived from extracted protocol signals and abstract evidence.

Rater population

Not reported

Signals refreshed

Aug 11, 2026

Open arXiv Find implementation Shortlist (0) RSS feed

Tags

Automatic Metrics Demonstrations General

Should you rely on this paper?

This paper has strong direct human-feedback and evaluation protocol signal and is suitable as a primary eval pipeline reference.

Use this as a practical starting point for protocol research, then validate against the original paper.

Best use

Primary benchmark and eval reference

Use if you need

A concrete protocol example with enough signal to inform rater workflow design.

What to verify

Validate the exact study setup in the full paper before operational use.

Main weakness

No major weakness surfaced.

Human feedback signal
Detected
From extracted signals
Evaluation signal
Detected
Eval setup described
Usefulness for eval research
75/100
High-confidence candidate

Use this as a primary source when designing or comparing eval protocols.

If you are doing eval pipeline work, start here

Human Eval Papers LLM-as-Judge Papers Pairwise Preference Papers Tool-Use Evaluation Search

Abstract

Self-consistency (SC) via majority vote is a widely used way to spend inference-time compute: sample N chains of thought, return the plurality answer. On the full GPQA Diamond benchmark (198 graduate-level science questions), majority voting reduces per-problem accuracy on a majority of problems for two instruction-tuned models from different families: 56.6% of problems for Qwen2.5-7B and 65.7% for Llama-3-8B, with Qwen the primary demonstration and Llama corroborating the direction from a near-chance baseline. The effect was pre-registered on a 151-problem confirmatory split after being observed on 47 exploratory problems, and all four confirmatory hypotheses passed. A grid oracle that routes each problem to the best N across {1, 2, 4, 8, 16, 32, 64} marks a theoretical upper bound 14 accuracy points above N = 1 for Qwen and 17 for Llama, an oracle bound requiring ground truth rather than a deployable method. No verifier-free gate reaches it: neither a plurality-agreement gate nor a token-entropy gate moves accuracy more than 0.002 from fixed-budget voting at N = 64. The mechanism is direct: confidence does not track correctness on these problems. In the highest-agreement bin the plurality answer is correct about half the time for Qwen, and for Llama that bin is less accurate than its lowest-agreement bin. We pre-register and confirm these findings on small instruction-tuned models; we do not test reasoning-native models, which we flag as the central open question.

What we could verify

These are the protocol signals we could actually recover from the available paper metadata. Use them to decide whether this paper is worth deeper reading.

Human Feedback Types

strong

Demonstrations

Directly usable for protocol triage.

"Self-consistency (SC) via majority vote is a widely used way to spend inference-time compute: sample N chains of thought, return the plurality answer."

Evaluation Modes

strong

Automatic Metrics

Includes extracted eval setup.

"Self-consistency (SC) via majority vote is a widely used way to spend inference-time compute: sample N chains of thought, return the plurality answer."

Quality Controls

strong

Adjudication

Calibration/adjudication style controls detected.

"Self-consistency (SC) via majority vote is a widely used way to spend inference-time compute: sample N chains of thought, return the plurality answer."

Benchmarks / Datasets

strong

GPQA

Useful for quick benchmark comparison.

"On the full GPQA Diamond benchmark (198 graduate-level science questions), majority voting reduces per-problem accuracy on a majority of problems for two instruction-tuned models from different families: 56.6% of problems for Qwen2.5-7B and 65.7% for Llama-3-8B, with Qwen the primary demonstration and Llama corroborating the direction from a near-chance baseline."

Reported Metrics

strong

Accuracy

Useful for evaluation criteria comparison.

"On the full GPQA Diamond benchmark (198 graduate-level science questions), majority voting reduces per-problem accuracy on a majority of problems for two instruction-tuned models from different families: 56.6% of problems for Qwen2.5-7B and 65.7% for Llama-3-8B, with Qwen the primary demonstration and Llama corroborating the direction from a near-chance baseline."

Benchmarks and datasets

GPQA

Reported metrics

accuracy
Human feedback details
Uses human feedback
Yes
Feedback types
Demonstrations
Rater population
Not reported
Expertise required
General
Evaluation details
Evaluation modes
Automatic Metrics
Agentic eval
None
Quality controls
Adjudication
Evidence quality
High
Use this page as
Primary benchmark and eval reference

Research brief

Metadata summary

Self-consistency (SC) via majority vote is a widely used way to spend inference-time compute: sample N chains of thought, return the plurality answer.

Based on abstract + metadata only. Check the source paper before making high-confidence protocol decisions.

Key takeaways

  • Self-consistency (SC) via majority vote is a widely used way to spend inference-time compute: sample N chains of thought, return the plurality answer.
  • On the full GPQA Diamond benchmark (198 graduate-level science questions), majority voting reduces per-problem accuracy on a majority of problems for two instruction-tuned models from different families: 56.6% of problems for Qwen2.5-7B and 65.7% for Llama-3-8B, with Qwen the primary demonstration and Llama corroborating the direction from a near-chance baseline.
  • The effect was pre-registered on a 151-problem confirmatory split after being observed on 47 exploratory problems, and all four confirmatory hypotheses passed.

Researcher actions

  • Compare this paper against nearby papers in the same arXiv category before using it for protocol decisions.
  • Validate inferred eval signals (Automatic metrics) against the full paper.
  • Use related-paper links to find stronger protocol-specific references.

Caveats

  • Generated from abstract + metadata only; no PDF parsing.
  • Signals below are heuristic and may miss details reported outside the abstract.

Recommended queries

Human demonstrations evaluation Accuracy

Contribution summary

  • On the full GPQA Diamond benchmark (198 graduate-level science questions), majority voting reduces per-problem accuracy on a majority of problems for two instruction-tuned models from different families: 56.6% of problems for Qwen2.5-7B and…
  • A grid oracle that routes each problem to the best N across {1, 2, 4, 8, 16, 32, 64} marks a theoretical upper bound 14 accuracy points above N = 1 for Qwen and 17 for Llama, an oracle bound requiring ground truth rather than a deployable…
  • No verifier-free gate reaches it: neither a plurality-agreement gate nor a token-entropy gate moves accuracy more than 0.002 from fixed-budget voting at N = 64.

Why it matters for eval

  • On the full GPQA Diamond benchmark (198 graduate-level science questions), majority voting reduces per-problem accuracy on a majority of problems for two instruction-tuned models from different families: 56.6% of problems for Qwen2.5-7B and…

Researcher checklist

  • Human feedback protocol is explicit

    Detected: Demonstrations

  • Evaluation mode is explicit

    Detected: Automatic Metrics

  • Quality control reporting appears

    Detected: Adjudication

  • Benchmark or dataset anchors are present

    Detected: GPQA

  • Metric reporting is present

    Detected: accuracy