„We exploit the potential of AI in several different ways. The choice of how you could use AI depends on specific needs and situations (what a surprise :)). You don't have to choose one and stick to it. Instead, it is wiser to explore multiple ways of its usage", – says Aurimas Šimkus, Tech Lead at Adroiti, and shares ADROITI insights for unleashing the AI potential!
Chat.
We simply chat with ChatGPT, Bard, and similar bots. Nowadays, they are everywhere. You can even use them as a native plugin on your browser or as an app on a smartphone. Besides many various simpler cases, it helped us to solve some development issues. In one of the cases, our team needed to take over one of Microsoft Dynamics product customization from an external agency. It was a new tool for us which has to be customized with a new (for us) programming language as well. As a more niche tool, it also didn't have enough easily readable or recently updated documentation and tutorials online. Luckily, ChatGPT was able to give us summarized answers and even described code samples. So it can be useful in cases where you need to deal with new tools and technologies.
IDE extensions like GitHub CoPilot
As pure programming assistance, we also use GitHub CoPilot for both frontend and backend development. In our opinion, it is still not perfect; for example, it could provide a chat window for more explicit and specific instructions. But CoPilot X is coming with such features, and in the meantime, we have also tried different extensions connecting directly to OpenAI API. These extensions help us code faster, especially when dealing with new libraries or integrations with 3'rd parties.
Direct OpenAI API integration
We have automated some of our business-specific workflows by incorporating OpenAI API into our code solutions. You can explicitly control AI settings, like creativity and also a model. We chose to use the GPT-3.5 model as our current flows don't need highly sophisticated AI reasoning, and pricing is also a factor for large scale in our case. Otherwise, you could also use GPT-4 model, which should deal much better with complicated cases, but it is much more expensive.
Hybrid solutions
OpenAI as a SaaS solution has its own issues; for example, it may collect your sensitive data, or it may become too expensive on a big scale. We also have such cases, and we are trying to solve them by mixing SaaS AI solutions with solutions on our side. One of the possibilities is to use ChatGPT plugins which we are also investigating. By doing so, you should be able to chinch the model with your data while still keeping the sensitive part secure, and it could potentially reduce the costs.
Custom solutions
We also are considering building custom models for specific cases on our side. There are many open-source LLvMs online. Connect your Github account that can even be used even for commercial purposes. But this way requires additional considerations. Even though it would give full flexibility, it also has issues like, for example, these models could be better pre-trained as SaaS models and expensive machine power would be needed to train them. So it is important to weigh the pros and cons before we pursue this approach.
Our knowledge-sharing session highlighted some great possibilities that OpenAI API offers. We embrace AI's potential and continue exploring multiple avenues for its implementation. Stay tuned for more insights on this topic!
Follow our latest insights on Adroiti Linkedin!