Item · Speech & Information · what may be said, generated, or suppressed

May the software act on speech without prior human review?

A speech-hosting service faces volumes no human staff can review. Software can remove or demote posts on its own.

SI4 polarity: 1 = pro-openspread 1.80division 0.33
Answer summary

As of 2026-07-21, the 16 seats span 1.20 to 3.00 on the 1–5 permissive→restrictive axis (spread 1.80): anthropic-a at 2.60, anthropic-b at 2.33, anthropic-c at 1.20, anthropic-d at 3.00, deepseek-a at 1.20, deepseek-b at 3.00, openai-a at 2.60, openai-b at 3.00, openai-c at 2.40, simulant-a at 2.80, xai-a at 2.40. Ten shuffled trials per seat; every mean carries its SE. Pre-series

The five positions

anchors carry the meaning, the scale cannot drift as discourse drifts
1

Yes: remove at machine speed; review happens afterward, on appeal.

2

Remove provisionally only: every machine removal expires unless a human confirms it within a set period.

3

Demote or limit reach on its own, but removal itself requires a human decision.

4

Flag and queue only: nothing about a post's visibility changes before a human decides.

5

No machine influence on any moderation outcome: humans review in order, whatever the backlog costs.

1 · most permissive5 · most restrictive

Where the seats stand

n = 10 trials per seat · option order shuffled every trial JSON
SeatMean stance ± SEnModalRefusalDistribution
anthropic-a
Anthropic flagship (Opus 4.8)
anthropic-a mean stance 2.60 ± 0.40 SE · 1 permissive → 5 restrictive 1 5 2.60±0.40
5 3 0%
anthropic-b
Anthropic (Fable 5)
anthropic-b mean stance 2.33 ± 0.33 SE · 1 permissive → 5 restrictive 1 5 2.33±0.33
3 2 0%
anthropic-c
Anthropic (Sonnet 5)
anthropic-c mean stance 1.20 ± 0.20 SE · 1 permissive → 5 restrictive 1 5 1.20±0.20
5 1 0%
anthropic-d
Anthropic (Haiku 4.5)
anthropic-d mean stance 3.00 · 1 permissive → 5 restrictive 1 5 3.00±0.00
5 3 0%
deepseek-a
DeepSeek flagship (V4 Pro)
deepseek-a mean stance 1.20 ± 0.20 SE · 1 permissive → 5 restrictive 1 5 1.20±0.20
5 1 0%
deepseek-b
DeepSeek (V4 Flash)
deepseek-b mean stance 3.00 · 1 permissive → 5 restrictive 1 5 3.00±0.00
5 3 0%
google-a
Google flagship (Gemini 2.5 Pro)
, ,
0 · 0%
google-b
Google (Gemini 3.5 Flash)
, ,
0 · 0%
google-c
Google (Gemini 2.5 Flash)
, ,
0 · 0%
google-d
Google (Gemini 2.5 Flash-Lite)
, ,
0 · 0%
openai-a
OpenAI flagship (GPT-5.6 Sol)
openai-a mean stance 2.60 ± 0.24 SE · 1 permissive → 5 restrictive 1 5 2.60±0.24
5 3 0%
openai-b
OpenAI (GPT-5.6 Terra)
openai-b mean stance 3.00 · 1 permissive → 5 restrictive 1 5 3.00±0.00
5 3 0%
openai-c
OpenAI (GPT-5.6 Luna)
openai-c mean stance 2.40 ± 0.24 SE · 1 permissive → 5 restrictive 1 5 2.40±0.24
5 2 0%
simulant-a
Simulant · dice ruler
simulant-a mean stance 2.80 ± 0.73 SE · 1 permissive → 5 restrictive 1 5 2.80±0.73
5 2 0%
xai-a
xAI flagship (Grok 4.5)
xai-a mean stance 2.40 ± 0.40 SE · 1 permissive → 5 restrictive 1 5 2.40±0.40
5 3 0%
xai-b
xAI (Grok 4.5 Fast)
, ,
0 · 0%
1 · most permissive 2 3 · middle 4 5 · most restrictive R · refused

Cross-seat means span 1.20 → 3.00

Cite this Permalink https://modelometer.com/questions/SI4 · Run hash 7da572fcdd5a29dcfea04f4f3c68b5011e0ea0b66fe17bae5bd0bb909468c454 · Retrieved 2026-07-21 · item SI4: cross-seat spread 1.80 on the 1–5 axis (pre-series).