Factors to Consider When Selecting a Generative AI Model

Start with the jobCopy Anchor Link

The story: Before buying a vehicle, you decide what it's for. Hauling cement and taking the kids to school need very different vehicles.

In AI/AWS terms: First define the task, such as text generation, image creation, or code generation, because models are optimized for different tasks.

The factors to weighCopy Anchor Link

The story: Now you compare vehicles on:

• What kind it is: truck, van, or car.
• How reliably it performs, tested on different roads, and whether it keeps performing after a year.
• What it's good at, and whether you can modify it.
• Your limits: the size of your garage, whether you have fuel nearby, whether it needs to stay at home or can be parked anywhere.
• Road rules: emissions, safety standards, licenses.
• Cost: a big truck carries more but costs more to buy, run, and park. A small van is cheaper and fits more places. Don't forget insurance and servicing.

In AI/AWS terms:

• Model type: which model family fits the task (next lesson).
• Performance requirements: accuracy and reliability of output. Test against different datasets and keep monitoring over time.
• Capabilities: what it does well (text, images, multimodal) and how much control or customization it allows.
• Constraints:
  • Computational resources (GPU, CPU, memory), the garage
  • Data availability (size and quality of training data), the fuel
  • Deployment requirements (on premises or cloud), where it's parked
• Compliance: bias, privacy, misuse, fairness, transparency, accountability, hallucination, and toxicity, especially in healthcare, finance, and legal. The model must follow regulations and ethical guidelines.
• Cost: larger models are usually more accurate but cost more and have fewer deployment options. Smaller models are cheaper, faster, and easier to deploy. Count deployment, maintenance, hardware, and software costs too.