There’s no hotter topic in tech right now than Generative AI, but what is it, why is it important, and how should you consider using it?

AI: a very brief history

Artificial Intelligence, or AI, has been a field of study for almost a century, with computer scientists like Alan Turing speculating on the possibility of creating machines that can think as early as the 1940s.

It took until the 90s for AI to become sophisticated enough to start solving real problems, aided by the increase in computing power available at that time, in fields such as speech recognition, spam filtering, and spanking Gary Kasparov at chess.

The 2010s saw the invention of deep learning, a type of artificial intelligence that uses many layers of neural network to process information. Unlike earlier types of AI, deep learning is much better able to ingest unstructured data, making it easier to quickly get value from whatever data you have lying around.

It’s still used today for image recognition, natural language processing (including sentiment analysis), fraud detection, and recommendation systems like the one that Amazon uses on its ecommerce site.

The present: Generative AI

And so to the present day, and Generative AI. These are deep learning models trained on huge amounts of data.

Generative models have been used for a number of years, in applications such as speech recognition and text classification. They attempt to “understand” the structure of the data set they’re given. From there, they can generate probabilistically “similar” data when prompted.

When combined with deep learning, and a large data set, these generative models become much more powerful. Today, they can be used to create text, images, audio, and video. That they can do this at such a surprising level of quality is why Generative AI has captured so many imaginations in the tech industry and beyond.

Broadly speaking there are two places you, as a CTO, should be considering making use of Generative AI tools: either as part of your product offering, or behind the scenes, to help you build your platform.

Using Generative AI to write software

Even when building entirely novel products, very little of the code we write from day to day is particularly unique. Generative AI tools, such as Amazon Q Developer, have been trained on large numbers of software repositories, and these can help reduce the time developers spend on that sort of code.

Plugged into your text editor, Q makes code suggestions as you type. This is a little like autocomplete, but where autocomplete can guess what you mean based on context such as the name of the function you’re implementing, or from a short comment describing what that block of code ought to do.

Developers all over the world are using these tools to deliver code more quickly. If you adopt this in your team (and you should at least be considering it), bear in mind that your humans are still responsible for quality control.

Using Generative AI models in your SaaS product features

Perhaps more interesting than code generation is the prospect of adding GenAI features to your own product.

You might consider exposing a chatbot to your users, enhanced with access to your documentation or knowledge base via Retrieval Augmented Generation (RAG) to help answer questions about how your product works, reducing the burden on your support team.

Some of the SaaS products I use have started inviting me to use AI to either create or edit written content. If your SaaS involves creating or manipulating text (perhaps you develop a marketing tool, or helpdesk software), you have the option to try this too. (So far the results I’ve seen are a mixed bag in terms of quality, but I imagine that this will improve as models get better).

If your product provides some kind of dashboard or other tool for manipulating data, using an LLM to take natural language input to drive the query engine can lower the bar for non-technical users to get value from their data. Honeycomb provided the first example I saw of this.

Generative AI on AWS

Amazon Bedrock provides access to a broad range of foundation models for you to build your AI features on, including the Claude 3 models from Anthropic, and Meta’s Llama models. If you’re building your product on AWS, you should consider adopting Bedrock for your GenAI features.

Each model has different strengths and weaknesses, and different pricing. Bedrock’s evaluation tools let you experiment with different models and choose the right one for your use case. You should expect to keep evaluating models: this is a fast moving field of technology research, with new models being released all the time. The model you choose today may not be the one you’re using next year.

Because Amazon Bedrock is an AWS service, you won’t need to send your customers’ data outside your AWS environment to use it, and AWS guarantees that your data won’t be used to train any of the models provided by the service.

Conclusion

Generative AI is a fast moving field. If you’re building SaaS on AWS, you should be evaluating Amazon’s flexible and secure offerings in this space.

Ready to start building your GenAI supported applications on AWS? Book a free chat to find out how we can help.

---

This blog is written exclusively by The Scale Factory team. We do not accept external contributions.