Evidence · first-party tested
Expression accuracy is a hard ceiling: across two different reference images, intense interrogation prompts still produced flat, neutral-to-unsmiling faces instead of the requested angry/guarded intensity, and the report rates this 4/10.
The observation
Expression accuracy is a hard ceiling: across two different reference images, intense interrogation prompts still produced flat, neutral-to-unsmiling faces instead of the requested angry/guarded intensity, and the report rates this 4/10.
Criterion: Expression accuracy ·
Query this
get_evidence({
tool: "scenario"
})MCP · mcp.aidemos.com/api/mcp
Free with attribution.
Free with attribution.
Real outputs, no retouching · every cell queryable via API & MCP · aidemos.com