Elephant and Monkey - Issues in generated AI

Using this story again in different context. I tried to generate images for elephant and monkey for this article. Generated AI using DOLLY model generated images. Below are images, issues and solution

Generated picture elephant and monkey
Generated photo of monkey and elephant
Generated image of monkey, elephant

Problem is - Monkey has trunk in generated images. The generated data violate integrity that monkey do not have trunk. In context of this story it makes it unusable. I want to show agile monkey and heavy elephant.
In software one can test, path code will take. It become extremely difficult to determine path /activiation neural network will have. Generative AI comprehend existing data and use other trajectories data can take. It can generate data that may be good for fun, creativity. But many times it may generate data that is unnatural. Using such data may maythat may have damaging immpact.
Generative AI provide great utility. However one need to have mechanism to determine whether generated content is violating assumptions/policies/laws.
This is reason we build compliance around generative AI. We have put together knobs to control which type of data should be geenrated vs which type of data should not be geenrated.

Problem in generated AI images

In this story I am trying to highlight how elephant and monkey are different. How elephant size and trunk is strength.

Trunk for Monkey

Adding a trunk for Monkey spoil the purpose. It spoil the purpose of analogy.

The key thing to note is one require some control while generating data. In software one can control and test path code will take. It become extremely difficult to determine path /activiation neural network will have. Generative AI comprehend existing data and use other trajectories data can take. It can generate data that may not make sense. Generative AI provide great utility. However one need to have mechanism to determine whether generated content is violating assumptions/policies/laws.


At dataknobs we help you

  • Create and build dataset
  • Generate data set
  • Identify compliance issues in generated dataset.
  • At Dataknobs we solve

    Generative AI comprehend existing data and use other trajectories data can take. It can generate data that may not make sense. Generative AI provide great utility. However one need to have mechanism to determine whether generated content is violating assumptions/policies/laws.

  • Create dataset using knobs to specify and control
  • Identify compliance issue in generated and real data
  • Identify blind spot in data, AI and generate test cases.<