Part 3/8:
The effectiveness of shotgunning is underscored by impressive statistics. Testing revealed an 89% effectiveness on GPT-4 and 78% on Claude 3.5 Sonet when subjected to 10,000 augmented prompts. However, its application is not limited to text—shotgunning also works seamlessly with audio and vision models.
For image-based models, augmentations are achieved by integrating typographic elements within visual prompts, adjusting their color, size, and position. Audio models demand a different strategy, where manipulating elements like speed, pitch, and background noise during vocal requests can yield successful outcomes. Here again, success rates are notable, with 56% effectiveness for GPT-4 vision and up to 72% for audio inputs.