Recently, artificial intelligence has made significant progress with the emergence of generative models*, as we’ve covered in our previous blog posts. These models represent a huge technological and market boom, offer a wide range of applications and provide surprisingly user-friendly solutions. But of course there are also limitations to these systems and their business use requires competence and experience.
Well, where else to look for competence and experience than in an artificial intelligence company. I asked my colleagues at Neuron Solutions how they use the new generative AI models in their work and gathered their insights on how these models should be used at the moment.
Otto Werschitz, Co-Founder and Head of Business Development at Neuron Solutions, relies primarily on the use of ChatGPT to clarify professional and business issues. “The interactive response feature eliminates the need to browse the web, but if something is missed, ChatGPT can be used to ask. However, the lack of resources can be confusing and may affect the credibility of the answer.”
Of course, ChatGPT has been tried by everyone in the company and many of us use it in our work. I, for example, don’t waste my time writing boring snippets of code, but have been writing them with the model recently. Gergő Ferenczy and Gergő Szabó reported that when they use a new package or API, they find it a quick way to learn about the relevant parts and functionality of the package by asking ChatGPT, as it often provides a very useful summary with code snippets. Gergő Ferenczy also uses it as research material, saving time by getting a good summary of the topic he is looking for and knowing what its possibilities are, he only needs to read up on the relevant topics in a targeted way.
In addition, Gergő Szabó likes to draw inspiration for his own hobby projects with the DALL-E image generating model, while Ádám Franyó and Noémi Vorák used the same model to create Christmas greeting cards for our clients. Meanwhile, Miklós Tóth uses another text-based image generation model, Midjourney, for the most important purpose: creating fairy tales for his little daughter.
But the list of models does not end there. Levente Szabados, scientific advisor and co-founder of our company, uses the Galactica AI model to search for relevant research articles. Ádám has also tried GitHub’s Copilot, a model that generates coding suggestions based on natural language commands, while others have explored the OpenAI Codex. And András Simonyi used a close relative of ChatGPT, a version of OpenAI InstructGPT called text-davinci-003, in a project.
And the list goes on, many of our colleagues (and customers) are learning about or using these new models, which, as the above shows, we find very useful and exciting, but we also caution business users to be careful, as there are still many limitations of the models. These models are still in test mode, so they are subject to error and require experience and competence to use properly.
Gyula Kovács, co-founder and managing director of Neuron Solutions, said that despite appearances, the use of these models is not self-evident. To use them effectively, it is important to understand how to properly incentivise these models to achieve the desired results. We call prompt engineering the method of using these models effectively. Prompting is simply the process of giving specific instructions to artificial intelligence models to produce a particular output. This can involve feeding the appropriate datasets, selecting the right instructions, and refining the outputs to match the desired outcomes.
The potential applications of new AI models are endless and we are excited to see how they will continue to influence our work in the future. Keep an eye on what’s new, because we’re sure it will change the way we all work significantly, and it’s important to stay up to date. Follow us so you don’t miss out on news from the world of artificial intelligence!
* Generative AI (GenAI) is the part of artificial intelligence that can generate all kinds of data, including voice, code, images, text, simulations, 3D objects, videos, etc. It draws inspiration from existing data, but also generates new and unexpected outputs, opening up new avenues in product design, art and many other fields.