People hear voices in ambiguous audio, and a “paranormal” framing makes it worse.
Whisper has been documented inventing words from silence and noise. The audio you hear is split-second slices of whatever is on air, plus noise. A software-defined radio dongle plus sweep software is the other path.
Key takeaways
- Don’t build anything that transmits.
- Their data was speech from people with aphasia, so it isn’t a noise test.
- Different audio, different models, different measures.
Condition C: synthetic white noise, no radio at all.
Read the full article: https://www.neuralgrimoire.com/build-an-ai-spirit-box/
Top comments (0)