Two Uncertainties Are Not One: Disagreement versus Decision-Risk in Cheap Structured Content Moderation
arXiv:2610.33258v1 Announce Type: new Abstract: Content moderation compresses a graded, contested human judgment into a binary act (leave up / take down) over billions of items, at a scale where running a frontier LLM on every post is uneconomical. A new class of frozen "System One" models returns typed, calibrated decisions instead of text at ~USD 0.000025 per call and ~100 ms. We run the first independent audit of one such model on 142,658 human-labeled comments, and benchmark it against 14 fr
Read original source
Source record
Research analysis
Research claim
Paper identifier: 2610.33258. Source account: arXiv:2610.33258v1 Announce Type: new Abstract: Content moderation compresses a graded, contested human judgment into a binary act (leave up / take down) over billions of items, at a scale where running a frontier LLM on every post is uneconomical. A new class of frozen "System One" models returns typed, calibrated decisions instead of text at ~USD 0.000025 per call and ~100 ms. We run the first independent audit of one such model on 142,658 human-labeled comments, and benchmark it against 14 fr
Evaluation record
Recent records to compare: “Memorilla: Latent Semantic Memory for LLMs” (Hacker News, Nov 6, 2026); “Show HN: Ru-Jev – an open model that scores caller-supplied options on CPU (MIT)” (Show HN, Nov 6, 2026). Shared themes are navigation hints; they have not been verified as the same need or independent corroboration.
What the result supports
Interpret results only for the reported model version, task, data and inference budget. Community votes, downloads or a benchmark position do not establish reliability in a different workflow.
Validity gaps
Check sample selection, baseline configuration, uncertainty, contamination risk, negative results and access to code or weights. No reproduction by WeWorkBuddy is recorded here.
Verification plan
Read the paper or model card and its evaluation appendix. Reproduce one bounded result against the stated baseline, then test a representative user task and record where the conclusion fails.