2024-12-22 [ENG]: How I use GenAI at work: ChatGPT

Generative AI continues to reshape the way we work and interact, particularly in creative domains. Predicting the full scope of its impact remains an open question, but adopting these tools is just a matter of time.

In this post, I’ll share my experience with ChatGPT, probably the most known AI assistant. It is a general purpose tool, ane has a wide set of applcations, but here I will list the top 5 in my daily work in research, teaching and PR.

What is ChatGPT?

ChatGPT https://chatgpt.com/ is an AI-powered conversational assistant designed to assist with a wide range of tasks, from answering questions and providing detailed explanations to generating creative content like stories, code, or academic essays. With its natural language processing capabilities, it allows users to interact in a conversational manner, making complex information accessible and engaging.

The first public release of ChatGPT version 3.5 took place on November 30, 2022. It was a version in the form of an interactive tool, which quickly gained great popularity and became a kind of symbol for interactive assistants.

In essence, ChatGPT acts as a virtual assistant that simplifies problem-solving, brainstorming, and content creation without requiring extensive technical knowledge. In my opinion, this technology has immense potential across industries like education, research, and customer service, offering innovative ways to engage with knowledge.

What I’m doing with ChatGPT?

The following description applies to a paid subscription.

ChatGPT is an assistant to which you can delegate any task. Since it has had a memory mechanism for several months, so you can also have it summarize the tasks I have used it for in the last year. Here is that summary

ChatGPT is a great tool for brainstorming section titles, acronyms, alternative names. It is great in giving a wide range of possible oslutions in tasks like “propose 20 different ideas for combining XAI with challenges in public administration. For each application, suggest a short, attractive acronym that starts with the letter A.” Usually, 50% of the suggestions don’t make much sense, and another 30% are completely unusable, but the remaining ones often help in finding an interesting solution.

ChatGPT can be useful for simulating a discussion partner when reviewing scientific paper, a presentation, or a specific topic. The assistant can suggest a list of questions for particular parts of the text and propose language adjustments. Such interaction can also be helpful if, based on an article, you want to create a list of points to cover in a class to introduce a new topic.

ChatGPT is quite good at translating between languages. Until now, I’ve been using DeepL for translations, which is a fantastic tool for translating between multiple languages. However, I’ve been increasingly turning to ChatGPT for translations, mainly because it’s easier to give the text a unified tone—make it more formal, slightly shorten it, use simpler words, etc. Overall, the only feature I really miss compared to DeepL is the ability to select synonyms with a single click. But other than that, ChatGPT is an excellent translator that often requires only minor adjustments afterward.

ChatGPT is a good assistant for writing scripts in languages I rarely use (e.g. bash, javascript, json formatting, xml, etc.).

ChatGPT can also work well in suggesting team-building activities. E.g., find a pub in Warsaw to celebrate the publication of an article on XAI in space applications. Or propose a networking activity, warming up discussions among experts working on AI applications in academia.

What I’m NOT doing with ChatGPT?

So far, there are few things I don’t do with ChatGPT. The list could be probably much longer, but as the tool gets better maybe some of these things will be possible.

Tasks that require critical thought as to whether they can be done. Currently, chatgpt is not able to judge what it can do and what it can’t. For example, for “give and 5 random numbers” will do something, but they are not always random numbers (it is not clear from what range).

Generate a academic bibliography or sources for less obvious facts. While he’s getting better at it, you should always check your sources.

Reviewing other people’s papers, theses, or any other content that is not publicly available. Admittedly, commercial licenses have an exclusion that interaction data is not used to train the model, but it is, among other things, stored in local memory. Besides, many journals explicitly write to reviewers that they cannot use GenAI tools in a review.

And how do you use ChatGPT?

Let me know in the comments.