Stay informed
Subscribe to our newsletter, our monthly look at food and non-food quality management.
ChatGPT is hot! You may have seen the tool popping up in the news a lot lately. On the one hand, complaining reports about students allegedly using ChatGPT to get their homework done. As a result, teachers would not be able to properly check whether the homework was done by the student or by ChatGPT. There were even disturbing reports that ChatGPT would take over people's work. On the other hand, there were also positive reports that ChatGPT could help reduce workload by providing support with tasks such as administration or email writing.
ChatGPT is an advanced chatbot: a program driven by artificial intelligence (AI). ChatGPT acquires and processes language in a natural way, actually just like a human does: the chatbot learns by storing all the information it receives from its users in its algorithm. In this way, it acquires an ever-increasing wealth of information and can also interpret users' questions more and more accurately and provide more useful answers to them.
For this blog, we explored some of ChatGPT's capabilities in learning and development. What can ChatGPT do and how useful is this chatbot for your work?
ChatGPT can be used for customer service purposes, such as answering frequently asked questions and providing solutions to problems that have occurred before. This can save a company's employees a lot of time, and the customer is also likely to be helped more quickly to a solution to his or her problem. After all, customers do not have to wait before getting a person on the line, but are helped immediately by the chatbot.
Below is an example of a ChatGPT plug-in for Moodle. For example, the chatbot can quickly answer common customer questions about Moodle. Note that ChatGPT will always generate an answer, even if it does not have sufficient knowledge about what is being asked of it. Answers may appear somewhat strange or anomalous in that case.
ChatGPT can analyze data from customers, giving companies insights into customer behavior, preferences and problems customers encounter. Based on this data, the chatbot can provide personalized product recommendations to customers. This allows customers to feel more valued by the company and increases customer loyalty.
One limitation of ChatGPT is that the chatbot has no emotions of its own and is unable to empathize with customers. Although the chatbot can give empathetic answers, it cannot properly assess certain types of emotional context of a situation. Thus, answers may appear empathetic, but do not actually connect well to the context and therefore still sound somewhat strange to the user.
In addition, ChatGPT is unable to provide recent information about companies or products, for example. This was evident, for example, when we asked the following question:
How satisfied are customers with Précon's e-learning training courses?
ChatGPT offers multi-language support and can also transfer pieces of text from one language to another, just like well-known translation tools Google Translate and DeepL. The difference with the latter is that you can configure the chatbot to your needs. For example, you can have the tone and style of the translated text adjusted. You can also choose to have the chatbot take into account cultural connotations and regional differences between languages. As a test run, we had a piece from our "Food Safety" training course translated into English below:
Translate "In addition to working food safe yourself, as colleagues, speak to each other accordingly. If your colleague is not working food safe, the product is not safe. By working food safe together, you strengthen the food safety culture in your company." to English.
"Besides ensuring that you work safely with food yourself, you also hold each other accountable as colleagues. If a colleague is not working in a food-safe manner, the product is not safe. By working together to ensure food safety, you strengthen the food safety culture within your company."
As you can see, the translation is very accurate. The words and terms used fit the context of food safety. The sentence structure and grammar are also correct.
And another piece of text, about taking measures during inspections, the chatbot translates as follows:
Translate "If during the inspection it turns out that the measure does not work properly, you are not sure if the product is safe. The finished product should then not be allowed to go to the customer and should be blocked." into English.
"If during the inspection it turns out that the measure is not working properly, you are not sure if the product is safe. The final product may not be sent to the customer and must be blocked."
Again, this is translated quite accurately. Do notice that "are you not sure if the product is safe" is translated very literally. The context of doubt is not reflected in the English translation. So it does remain important to go through the translation properly yourself.
Sometimes texts in training courses need to be made simpler, for example for employees who have difficulty reading and writing. Therefore, we asked the chatbot to generate the translations in simple language.
Translate "If the inspection shows that the measure does not work properly, you cannot be sure that the product is safe. The finished product may then not go to the customer and must be blocked." into simple English.
If it is found during the inspection that the measure is not working well, you are not sure if the product is safe. The final product may not be sent to the customer and should be blocked.
When we put both translations side by side, we see some minor differences. The word choice of the first sentence seems a bit simpler: "if it turns out that" has been replaced by "if it is found that" and "is not working properly" by "is not working well." So the chatbot also seems to be able to answer our more specific question.
ChatGPT can also be useful for developing a learning solution. For example, when writing a didactic concept with learning objectives. Since we at Précon often create training courses around the topic of food safety, we thought it would be interesting to ask the chatbot to create some learning objectives:
Can you create learning objectives for food safety training?
The learning objectives seem to have a logical structure. The answer begins with a learning objective about what food safety is and then moves to the risks, the rules set up by laws and regulations, and finally how measures should be applied in practice.
Note that some of these learning objectives are not easily measurable. For example, how do you measure whether someone has an understanding of critical control points? This is not clear from the learning objective.
You must be able to measure the outcome of what trainees have learned in a training course based on the learning objectives. Learning objectives must therefore be sufficiently specified. you can use, for example, Bloom's taxonomy. This is a list of verbs that help you specify learning objectives, such as 'the student can name X parts' or 'the student can explain concept X'.
Now we want to know if the chatbot gives more targeted answers if we make the question more specific:
Can you create learning objectives for food safety training at a meat processor?
Indeed, the proposed learning objectives seem to focus more on various aspects of food safety at meat processing plants, such as critical control (management) points in the meat processing process and preventing cross-contamination.
To know whether a student has mastered a learning objective, the student takes a test with test questions about each learning objective. We wondered if ChatGPT could also create test questions about food safety and asked the following question:
Can you create exam questions on food safety?
No strange suggestions, we think. Most of these questions also appear in our Food Safety (HACCP) training, albeit phrased differently.
For tests, we always use multiple-choice questions. Open questions where the student has to fill in something himself are difficult for an e-learning program and often require human input to assess the questions properly. We therefore made the question more specific: the questions must be multiple choice and deal with food safety in the food industry.
Take multiple-choice questions on food safety in the food industry
As you can see, the questions are now multiple-choice and cover various aspects of food safety in the food industry, such as possible contamination, the meaning of HACCP and listing allergens on packaging.
However, we do question the correct answer options. For example, question 2 on the correct temperature zone for storing perishable food states that it should be between 0 and 5 degrees Celsius, while in Food Safety (HACCP) we state that perishable food should be stored at a temperature of 7 degrees Celsius or below. So it is important to regularly compare information generated by ChatGPT with other reliable sources and verify it, to make sure the information generated is correct.
We'll make the question a little more specific and take the example of food safety at a cheese production facility:
Can you do multiple-choice questions about working food safely at a cheese production facility?
Answers:
The questions are now specifically about aspects of food safety at a cheese production facility. So specifying questions is important when getting started with ChatGPT: the more focused your question, the more focused the chatbot's response.
The answers ChatGPT provides are not from humans, but generated by the AI of a computerized chatbot. AI chatbots are not considered authors by law, so the information they generate cannot qualify for copyright protection in this regard either.
Exceptions are images or statements of existing persons generated by ChatGPT. This content is protected by copyright. But the problem here is that ChatGPT cannot always accurately indicate the source from which the given information comes. The chatbot is trained to extract pieces of information from a large pile of texts from different sources, such as Wikipedia, books and scientific articles. With these pieces of information, it then forms an answer.
Advice is therefore to regularly compare the information provided by ChatGPT with multiple sources and verify this information. And if answers generated by ChatGPT appear to contain statements or images from an existing author, it is best not to modify or use this content.
ChatGPT is not exactly energy efficient at the moment. The chatbot retrieves its data from data centers and these require a lot of power to keep running. Scientists have calculated that having ChatGPT run a training session would be equivalent to 500 tons of CO2 emissions, which is as much as 1,000 cars driving 1,000 kilometers each.
Several research bodies such as TU Delft are already looking at energy-efficient solutions for AI programs such as ChatGPT. For example, making programs search for data more effectively and use smaller amounts of data. Having data centers replaced by an efficient AI chip is also being explored as a possibility. Scientists are investigating whether they can have tasks such as speech recognition, image recognition and understanding what someone has written that are now performed by data centers performed by the algorithm in a chip. This algorithm would work much more efficiently than a data center, similar to how our brain's algorithm works.
Such power-efficient AI programs and chips are expected to appear on the market within a few years. So until then, we are left with power-hungry AI programs, including ChatGPT.
At first glance, ChatGPT seemed like a dangerous tool that could threaten our work. After we took a closer look at it and also tried it out, it turned out, on the contrary, that ChatGPT can help and support us well in various tasks: collecting customer preference data for marketing purposes, translating texts and requesting laws and regulations around food safety, for example, in performing these tasks ChatGPT is very fast and accurate.
ChatGPT has many advantages, but unfortunately also some disadvantages. For example, it can quickly gather customer preferences and respond to them, but cannot properly assess certain types of context of customer-specific situations and provide a logical answer to them. Also, empathy is often still lacking in answers given by the chatbot. In addition, data centers of ChatGPT and other AI programs consume a lot of power and the use of such AI programs is not environmentally friendly in that respect.
If you want to use ChatGPT, we recommend using the chatbot as a tool to gather information or as a source of inspiration for information. Be sure to make your questions specific: the more specific your question is, the more specific the chatbot's answer will be. Also make sure that you regularly compare information generated by the chatbot with other reliable sources to verify whether the information is correct. After all, the chatbot still sometimes wants to give incorrect information.
Contact us at info@precongroup.com or +31 (0)30 - 65 66 010.
Subscribe to our newsletter, our monthly look at food and non-food quality management.