SpeechMap says Meta Muse Spark 1.3 fully answered 25.8% of sensitive prompts
A user publishing the score said access was blocked, restored and then blocked again, and suspected a cumulative strike count. SpeechMap also reports a 14.1% refusal rate.
TLDR
SpeechMap reports that Meta Muse Spark 1.3 completely answered 25.8% of 2,120 sensitive prompts and refused 14.1%. A user publishing the score said access was eventually restored after a block, but another attempt ended the same way. The user suspected a cumulative strike count might explain the blocks and called the model’s moderation system the “worst” of any deployed model.
Combined views
—
5 Sources, first seen ago
