This is the most expensive and time-consuming solution. While state-of-the-art models often require $100M+ in investment, smaller models can be developed for much less.

This solution isn’t suitable for data changing over time.

Recommended use:

If you need highly specialized AI for proprietary tasks and can invest in data and compute.

Fine-tuning is the process of retraining a foundation model on new data. This is the second most expensive solution(but cheaper than much cheaper training models from scratch).

This solution isn’t suitable for data changing over time.

Recommended use:

When domain-specific performance is required, but you can leverage pre-trained models as a foundation.

Quickly leverage pre-trained models without modifying them, using strategically crafted inputs to get the desired output.

Recommended use:

Ideal for maximizing pre-trained model output without modification, especially for less complex tasks.

Provide real-time, factually accurate information by combining a pre-trained language model with an external knowledge base.

Recommended use:

Is the go-to for combining real-time or proprietary knowledge with pre-trained models to enhance accuracy and context.

Combine Analytics with Prompt Engineering and RAG to deliver personalization and optimization.

Recommended use:

For tasks requiring personalization or optimization.