*Generative Generative AI is best used when*:
1. The task requires understanding or generating human language, images, or complex media. The system needs to interact naturally with humans in a conversational or adaptable way.
2. The goal is to create something new (summaries, code, designs) rather than just categorize what already exists.
3. You don't have a perfectly labeled dataset, or the dataset size is only enough to test the system.
4. The process requires multi-step reasoning or following a chain of logic to reach a conclusion.
5. Speed to deployment is critical and can be achieved by prompting an existing model rather than building an architecture from scratch.
**Generative Generative AI shouldn’t be used when**:
1. Absolute factual accuracy is non-negotiable (e.g., calculating taxes or medical dosages where "hallucinations" are fatal).
2. The latency or cost of running a large model outweighs the value of the automation (e.g., using a massive LLM to do basic math or simple string formatting).
3. Sensitive data can’t be sent to the cloud and obfuscation/anonymization is not possible, no on-premise model, no on-premise LLM.
4. Deterministic outputs are a must