There's no such thing as the best GenAI tool
Tools like ChatGPT and Gemini are like multifunctional stand mixers or drills: good for most users, but not always suitable for professional use. Their effectiveness also always depends on the user and their purpose. And new GenAI tools are popping up like mushrooms, but which one is the best? That's what we asked Christophe Benoit (Erasmushogeschool Brussel).
Since late November 2022 – when ChatGPT 3.5 became available to the general public – an avalanche of new GenAI applications has been coming our way. During my many workshops and keynotes, and especially now that my book Zelf aan de slag met AI (Getting Started with AI) is being published, people very often ask me what the best GenAI applications are. Then I always mention multifunctional stand mixers or multifunctional drills. These appliances have many functions but none of which are really used in a professional setting. For the occasional, non-professional user, those appliances are very useful and more than adequate for the tasks that need to be performed
GenAI applications are similar: there are applications like a ChatGPT or Gemini that have a lot of features, especially the free versions. For most users, the capabilities offered by these standard applications are most certainly sufficient. But if you become adept at using these standard applications, you will find that for your specific way of working, there are limitations that can be remedied with another, more specialised application and with which you can work much more professionally.
Take image generation, for example: you could use DALL-E 3 and now Sora (built into ChatGPT-4o), but to truly create the image you had in mind, you are limited by the application’s capabilities. Midjourney, on the other hand, specialises in image generation. The tool offers the user a lot of image creation options – options that don’t exist in ChatGPT-4o. Midjourney is not free and requires a learning curve to use it properly. But its results are much better than ChatGPT-4o. In this view, you can consider ChatGPT (somewhat irreverently) the Swiss army knife of GenAI applications: it’ll get the job done, often with satisfactory results.
It is important to be aware of the capabilities and limitations of a tool, to be able to identify its strengths and weaknesses, so that you can also use the tool correctly. An example of this is the difference between ChatGPT and Perplexity; two seemingly identical applications that you can prompt and that give you an answer. But looks can be deceiving, as these two applications are fundamentally different. ChatGPT is an incredibly powerful writing engine while Perplexity is a powerful search engine (think: a Google on steroids). Once you know this, you will get better results. For instance, let’s take the prompt ‘What does an Applied Computer Science course at Erasmushogeschool Brussel cost?’ Perplexity will give you an accurate answer, while ChatGPT will give you a hallucination.
But the most important aspect of using GenAI applications remains the person using the application. Humans are still the biggest limitation to the possibilities of GenAI. We need to start thinking differently to reveal the true potential of GenAI applications. One thing to keep in mind is that GenAI applications like ChatGPT and Gemini are conversational models, which generate the best results by having a conversation and asking questions – just like talking to a colleague. For example, you can solve a problem incrementally by having a conversation. So it's best to ask for a step-by-step guide on how to bake a cake, instead of asking for a recipe for a cake. If you don’t understand a particular step, just keep asking about that step. This technique is called ‘chain-of-thought prompting’. You can also give input that you wouldn’t normally ask a colleague because it seems absurd. An example is the ‘self-criticism’ technique where you ask the tool to critique its own answer. An example: ‘Give me 3 good names for a vegetarian restaurant in Brussels and tell me why they are not good.’ This lets the application argue with itself.
Wat is de beste applicatie? Wel, in eerste instantie is dat die jou in jouw specifieke situatie het beste kan helpen. Gebruik je ze om vlug iets te genereren of gebruik je ze in een professionele context (dan ben je meestal ook beter af met een betalende versie)? In de tweede plaats is het belangrijk om de mogelijkheden en beperkingen te kennen van de gebruikte applicatie. Last but not least, is een opleiding “prompt engineering” zeker geen overbodige luxe.
Learn more
Prompt engineering: how to get the best result out of ChatGPT, Copilot or LeChat
Chatting with AI in everyday language doesn’t guarantee the right answer. That’s where prompt engineering helps. Only by understanding AI, crafting clear instructions, and refining them again and again, you can unlock useful results. Guillaume Vanhalst from Effixis explains the key principles, techniques, and the benefits of mastering this skill.
AI as a critical thinking coach
AI can make work and information retrieval faster and more efficient, but can it also promote critical thinking? The European research project TITAN is studying this by developing AI tools that encourage critical thinking and combat disinformation.
Is DeepSeek safe?
There are many safety concerns regarding DeepSeek: its apparent political censorship received a lot of attention in the media, but this can still be circumvented. A more pressing concern is privacy: DeepSeek’s online app offers no way to avoid having your information sent directly to China.
Christophe Benoit
Senior researcher at the Artificial Intelligence Knowledge Centre of Erasmus University College Brussels.
And more: AI Business Integrator, Microsoft Certified Data Strategist, Workflow Automation, Course Designer, Entrepreneur.
Learning & Development needs an enterpreneurial mindset to excel!
Research interests: prompt engineering advanced, AI implementation in the organisation, sentiment analysis, data analysis with GenAI
