ChatGPT represents a new height in the development of artificial intelligence and human technology. It has completely altered the way in which we acquire knowledge and elevated the writing and expression abilities of many. With its formidable language comprehension and text generation capabilities, ChatGPT has become a tool of productivity and creativity for humanity.
However, when I share ChatGPT with the general public, those without an AI background, the response I receive is not just one of praise and wonder but also of scepticism and doubt.
Some claim to have seen many chatbots before and that ChatGPT is just one among many. The chatbots of the past were unable to understand the intent behind human expression. But ChatGPT is not like them. Its powerful language comprehension and generation capabilities make it an indispensable tool for the future.
Others believe that artificial intelligence cannot surpass human wisdom, that it is merely learning from the data created by humans. Regardless of the truth, artificial intelligence does not need to surpass the creativity of all humanity to affect our daily lives. In fact, as long as artificial intelligence surpasses the capabilities of each individual, it can better assist people and bring a transformative impact to our work and way of life.
Still others claim that ChatGPT produces erroneous content, fabricates false information, or engages in seemingly logical but erroneous reasoning and that we need to stop using it. But doing so is like throwing out the baby with the bathwater. ChatGPT is indeed useful because it simplifies the process of finding information, as fact-checking is always easier than searching for information from scratch. Moreover, as artificial intelligence technology is increasingly being applied, we should become familiar with ChatGPT and other AI tools, understand in which areas they are prone to errors, and better avoid them rather than simply shunning them.
Yet others believe that using ChatGPT to automatically obtain answers impedes the development of critical thinking and problem-solving abilities. However, using ChatGPT does not mean delegating human thought to machines or stopping human thinking. ChatGPT cannot automatically provide profound insights or perceptive ideas. In fact, when it comes to uncommon questions, ChatGPT can only provide general and trite opinions. The key is for the user to provide the opinions and for ChatGPT to create a coherent article.
Another argument claims that the content created by ChatGPT is soulless as it lacks consciousness in its creation. However, the correctness of this argument is not important. The most important point is that artificial intelligence technology will undoubtedly become increasingly advanced in the future. Imagine the world twenty years from now, where artificial intelligence is a ubiquitous part of daily life. Looking back at ChatGPT, people may have different feelings. We must now familiarise ourselves with artificial intelligence technology because it is becoming more deeply integrated into our daily lives.
Now, let us take a closer look at the technology behind ChatGPT: natural language processing. Even within the NLP field itself, many people have not yet realised the disruptive impact of ChatGPT.
The evolution of research methodologies of natural language processing, from rule-based methods to machine learning to deep learning, is a testament to the complexity of human languages. The complexity determines that natural language processing is not as simple as writing a few rules. It requires constant improvement and exploration. Now, coreference resolution, dependency parsing, and part-of-speech tagging have all been discarded by natural language processing. In the future, even named entity recognition and grammatical error correction will seem comical. With ChatGPT revolutionising the NLP landscape, NLP researchers must actively familiarise themselves with this cutting-edge technology to embrace this new era.
Some researchers claim that ChatGPT is insufficient for certain widely researched tasks, such as machine translation. In fact, these researchers often use poor prompts and have not undergone proper dialogue engineering training. Additionally, the advantage of using ChatGPT for translation lies not so much in its accuracy compared to existing machine translation models, but in its ability to simultaneously improve the text, style, and tone. In these areas, there are no datasets to fine-tune language models, but general-purpose language models like ChatGPT can solve this problem.
Some researchers even view ChatGPT as nothing more than a blurry compression of the Internet. However, to view ChatGPT as a compression algorithm is an incomplete understanding, as ChatGPT has at least four distinct capabilities: generalised question answering (being able to solve other tasks after training on various tasks), text generation (predicting the next word based on previous words), knowledge base (storing knowledge in the model parameters), and chain-of-thought (explicit reasoning based on text).
It is a lamentable truth that many individuals, even those among the NLP community, lack an intuitive understanding of ChatGPT and the level of technological advancement we have achieved. They have not truly experienced ChatGPT, but instead judge this new technology based on a handful of trials or even their imagination. Therefore, it is of utmost importance to address these misunderstandings by offering a free ChatGPT API readily accessible to us all.
