The best AI-focused apps and services set themselves apart with curated data for training. When your go-to prompt engineering tricks can't squeeze out that last 20% of improvement, or when you need a smaller, cheaper, or faster model for a specific task or domain-focused use case, fine-tuning or custom pre-trained models are the only paths forward. But if you aren’t quite there yet, there are better ways to get sizable differences in quality from your prompts.
The Power of High-Quality Examples
Before considering fine-tuning, there's another approach worth exploring: adding high-quality examples to your prompt. This method can deliver significant performance improvements without the extensive commitment of fine-tuning.
The principle is straightforward: the more diverse, high-quality examples you include in your prompt, the better your results tend to be. It's like giving your model a cheat sheet of exactly what you want it to do.
So, how do you acquire a reasonably large set of top-notch examples? Creating them yourself is an option, but it's time-consuming and you are probably not a domain expert or an amazing writer and doing it yourself may not provide the diversity you need for optimal results.
Introducing Our Free Few-Shot Example Service
We're offering a free and simple way to get high-quality, diverse examples tailored to your specific use case: our new few-shot example service.
Here's how it works:
You provide us with your prompt and describe your desired output.
Our team of human domain experts crafts a free sample set of five diverse, high-quality examples that align with your needs.
You receive these examples over email, ready to be copy and pasted into your prompt.
With our few-shot example service, you can:
Quickly boost your model's performance
Explore the potential impact of more examples on your specific use case
Lay the groundwork for potential future fine-tuning projects
Looking Ahead: The Path to Fine-Tuning
While high-quality examples can take you far, some projects may eventually require fine-tuning to reach their full potential. Our few-shot example service is just the beginning. As you work with these examples and assess their impact, you'll gain valuable insights into your model's performance and your specific needs.
If you find yourself consistently bumping up against the limits of what examples alone can achieve, that's when fine-tuning comes into play. And guess what? We specialize in collecting high quality data for pre-training and fine-tuning using our meticulously curated set of domain experts.
Comments