AI TrackerFrontier desk · AEST
Back to news
normalaitech

Understanding Jev-Style System One Models: Does Structure Reduce Selection Bias?

arXiv cs.CL · 27 Oct 2026, 03:48 AEST

Selection bias in large language models persists despite efforts to mitigate it. A study investigates whether Jev-style decision structures can reduce this bias, revealing significant challenges in maintaining accuracy across varying presentations.

Key facts

Read aloud

Selection bias in large language models persists despite efforts to mitigate it. A study investigates whether Jev-style decision structures can reduce this bias, revealing significant challenges in maintaining accuracy across varying presentations. Selection bias in large language models L L M s can persist despite instructions to ignore presentation order or prompts that elicit reasoning. Across two Jev versions, twelve Jev reproductions, and six generative L L M s, we trace individual answers through eight presentations and compare them with fixed-order repetition. On our primary collection of one thousand seven hundred questions from two benchmarks, Jev's accuracy varies by less than two percentage points, yet its cumulative failure set expands by twenty seven--twenty nine percent relative to the original order. This loss of initial successes is substantially larger than under fixed-order repetition, even for highly repeatable Jev reproductions. These results expose a gap between repeatable decisions and decisions that remain correct across equivalent presentations.

Sources in this cluster