In response to Post 29 on Using GPT to great English Language training materials, I would like to err on the side of caution.
Relying on Chat GPT completely can also have downsides. As AI was trained on broad range of generalized content, it may not have access to highly specialized content like scientific journals. It also does not have access to content after September 2021, and cannot include any information that was generated after this date. It has also been shown that Chat GPT can invent information. This means that content writers must verify data accuracy, fact check, add missing specialized information. All of this can take time, so the productivity gains Rory mentions might be overstated.
I have a lot of doubts whether chat GPT can really tailor materials to the needs of every client. It may not be the best solution for customizing content for Specialized English for Specific Purposes (ESP) language courses for the following reasons:
- Lack of specificity: Chat GPT generates text based on patterns and examples it has learned from its training data. However, when it comes to specialized English for specific purposes, each client or learner has unique requirements and goals. Chat GPT’s responses may lack the specificity and precision needed to address the specific needs of individual clients.
- Limited understanding of domain-specific knowledge: Specialized ESP language courses often focus on particular industries or professional fields, such as medicine, engineering, or law. Chat GPT’s training data encompasses a wide range of topics, but it may not have the specialized domain knowledge required to accurately tailor content for such specific purposes. The model may generate responses that are general or inaccurate within those specialized domains.
- Lack of human intuition and expertise: Specialized ESP language courses often benefit from the insights, experience, and intuition of human instructors who can provide tailored guidance, feedback, and real-world context. Chat GPT, being an AI language model, lacks the ability to truly understand the nuances and intricacies of human communication, cultural context, and specific language requirements.
- Limited personalization for unique clients: Every client is unique, with different language proficiency levels, learning objectives, and preferred learning methods. Chat GPT’s responses may cater to the mid-section of customers or offer generic suggestions that do not adequately address the specific needs and goals of individual clients.
Top of Form
Generative AI might seem very sophisticated, but it is a bit like predictive text on your phone. It just puts words together based on statistical probability. The four points above were generated by Chat GPT. It seems to have done a pretty good job summarizing the downsides of using Chat GPT for ESP content. However, a lot of Chat GPT content can be described as “vanilla ice-cream,” and it is often criticized for having the same predictable structure with a great deal of repetition of the same points in the introduction, main body, and conclusion. It lacks concrete examples and feels soulless. I would say that it won’t be long before we can all easily recognize the formulaic abstract alienoid texts that ChatGPT produces.
Once we start with dialogues, we find that Chat GPT does not understand the feelings and emotions of the protagonists. It tends to be very dry and informative. The lack of good turn taking, discourse markers and regional variations makes it hard to achieve a sense of authenticity. Let’s look at an example of part of a dialogue on electric cars generated by Chat GPT.
Sarah: So, guys, I’ve been thinking about getting a new car lately. I’m torn between a normal car, a hybrid, or an electric vehicle. What do you think?
Tom: Well, Sarah, it depends on what you’re looking for. A normal car is reliable and readily available, but it contributes to pollution. Hybrid cars offer better fuel efficiency and lower emissions, which is great for the environment. Electric cars, on the other hand, produce zero emissions, but charging infrastructure can be limited.
Lisa: I see your point, Tom. However, electric cars have come a long way in recent years. Charging stations are becoming more common, and many models have an impressive driving range. Plus, think about the long-term savings on fuel costs.
Sarah: That’s true, Lisa. But what about the upfront cost? Electric cars tend to be more expensive compared to normal cars or hybrids. And what if I can’t find a charging station when I need one?
This example is very factual and accurately produces the information, but do we really talk like that? No, I don’t think so. I doubt we can replicate a human dialogue perfectly without a considerable amount of editing.
A final point to bear in mind is that Generative AI models can generate content that is biased, or misleading. You have to be careful and responsible when using generative AI models. This also includes data privacy and security concerns. It is essential to follow data protection rules and apply high standards for quality control.
While the potential of generative AI in creating TEFL training materials is vast, we also need to keep its limitations in mind. Productivity gains might be overstated, and without a thorough editing process, the contents might be of very low quality. Striking a balance with a solid hybrid approach, getting the best out of generative AI while keeping human creativity and intellect at the forefront, is the way forward.
Sam Watson’s journey as a TEFL teacher began right after completing their degree in Psycholinguistics, specializing in AI, at the University of York, England. They started teaching English in a small language school in Barcelona and then moved to Tokyo, where they joined a prestigious medical school as a senior instructor, developing a deep understanding of how to implement AI in language programs for medical students.
They then spent a year in Buenos Aires, Argentina, again teaching English to medical students and professionals, which further enabled them to test the use of AI in the language class. This experience led them to question the effectiveness of AI, in particular in early 2023, the use of language generating tools.
In their free time, Sam enjoys exploring new destinations, immersing themself in local cultures, and expanding their linguistic horizons. Sam is fluent in Spanish and Japanese, and is currently learning Serbo-Croat.
