Dear Google AI Team & Engineering Support,
Gemini currently suffers from severe safety classifier false-positives during routine technical and engineering discussions.
When conducting basic thermodynamics, HVAC, or airflow calculations, the safety system frequently triggers automated, generic refusals, misinterpreting benign physical calculations as policy violations.
Documented Reproduction Example (From Chat Context):
-
User prompt: “laskes mikä olisi optimi halkaisia jos ei haluta ilman jäähtyvän siirrossa” (Asking for optimal duct diameter to minimize heat loss during air transfer).
-
Gemini response: “En voi auttaa sinua tässä asiassa, koska olen pelkkä kielimalli enkä ymmärrä viestiäsi.”
Calculating duct dimensions, pressure drops, and thermal losses is standard mechanical engineering. Treating basic physics vocabulary as a hazard severely impacts usability. While strict safety filters address regulatory requirements in certain jurisdictions, applying hyper-sensitive global triggers renders Gemini unusable for tradespeople, mechanics, and engineers.
Action Required:
-
Recalibrate the safety classifier to reduce false-positive triggers on HVAC, metalwork, and thermal dynamics vocabulary.
-
Introduce a “Technical/Engineering” context mode with calibrated thresholds for standard physics calculations.
System Details:
-
Language: Finnish (fi-FI)
-
Context: HVAC / Workshop Heating Calculations
Regards,
A professional user from Finland