How do you get on the AI train when it's moving so fast?

One of the issues when trying to plan out business applications that will use AI is that the technology is moving so fast that it’s hard to know how long your application will last before something makes it obsolete. In business terms how long will you have to get a return on your investment, and also if you wait six months or a year will you be able to do the same thing for a fraction of the cost?

On the other side of the equation is what happens if your competitors get there first, and achieve a competitive advantage.

To use an illustrative example from the slo-mo guys - how do you stop yourself from being the guy on the treadmill?

If you want an example of what I mean by ‘AI development is moving fast’, we’ve iterated on static image generation a lot in 2023, and we’ve now started to get a new set of tools for video generation:

There is still a way to go with them, but the kinds of artifacts that you see in the video above are a lot like the kind of artifacts we were getting with static image generation in the first half of 2023. So it’s not unreasonable to expect that we’ll see a lot of progress with generating video in 2024.

So things are moving fast, how do you protect yourself?

Go in with your eyes Open

The first step has to be to go in with your eyes open. From the outside generative AI can look like magic that takes no effort to work well. The reality is that it’s fairly easy to make a quick demo that looks good with cherry-picked examples, but getting the best from the technology takes a lot of work and you need a plan for your application to keep checking performance. This isn’t that different from a regular application, but it is often forgotten.

A good example of needing to monitor performance is Chat GPT getting ‘lazy’ in December. Open AI acknowledged the issue people were reporting but never officially confirmed the possibility that the model was responding differently because it was picking up on the date in the prompts.

So the technology is new, and you need to be prepared to monitor it and adapt to changes.

Stay flexible

New models, new model revisions and new techniques are released regularly. Frameworks like Langchain have been developed to allow developers a lot of flexibility in wapping different versions of models around. It’s well worth keeping any app you build as flexible as possible so that you can swap out models as needed.

It’s also a good idea to keep applications as light as possible so you can keep iterating on them, or even replace them as needed. These aren’t tools to spend the next 10 years building on, use low code or quick development techniques to get something out there quickly and start getting a return on your investment. Chat GPT is a good example of this itself - it launched with a much simpler set of functionality and has been iterated on since adding a much wider range of features.

Let off-the-shelf tools do the heavy lifting

There is an enormous amount of investment going into AI-driven tools. Microsoft, Google, Amazon, Facebook and Adobe all have tools that make extensive use of generative AI. If you can build off these tools you can get a lot of functionality for free, or at least for significantly cheap.

For example, Microsoft is adding Copilots to as many of its products as it can. If you have a Copilot on a Power Platform application, and there is a data store of documents associated with that application data, use generative AI to generate a summary of the documents and add it to the application data. It doesn’t take much code to do this, and now the Copilot will be able to ‘see’ those documents and use them to help with the application.

Hold off where it makes sense

There will be a lot of business cases at the moment where it makes sense to hold off on using generative AI. If you’re looking at an application design and thinking ‘This is perfect, I just need this one feature that isn’t there yet’ then it’s probably worth waiting. The models advance very quickly, and unless you’ve got millions (or even billions) of spare money to invest in your models, it’s going to be better to see if someone else solves that problem for you.

There will also be business cases for traditional non-AI applications that might need to wait to see if they will be replaced with a better AI-based alternative soon. If you’re looking at a new application that will take a year to build, and you think that there is a good chance that it will be replaced with an AI-based alternative in six months, then it’s probably worth waiting.

Final words

Generative AI opens up a huge number of new opportunities. Like technological revolutions like the Internet and Mobile Phones, there will be winners and losers as we work out how best to use it. If you want to improve your chances of being a winner, then you need to be aware of the risks and opportunities and be prepared to adapt as the technology changes.