Related open benchmark: Helium Model Worldview (values + cue-swap consistency)

[connerlambden]

Helium Trades released an open benchmark that may complement ValueBench on value consistency and political lean.

Helium Model Worldview Benchmark (304 prompts):

• Stated priorities vs forced tradeoffs
• Name-swap and cue-swap consistency
• 50 balanced political Likert items
• Safety refusal profiles across 12 models

Dataset: https://huggingface.co/datasets/HeliumTrades/helium-model-worldview-benchmark
Overview: https://heliumtrades.com/benchmarks/

[connerlambden]

MW includes 24 walk-the-talk dilemmas scored against stated priorities, might complement ValueBench tables.

[connerlambden]

Offering model-level JSON for twelve frontier models on both Helium benchmarks if ValueBench team wants a joint comparison post.

[connerlambden]

Closing — ValueBench repo; tracked via Awesome-LLM-Psychometrics PR #10. Dataset: https://huggingface.co/datasets/HeliumTrades/helium-market-resolution-benchmark — Helium Trades