2. The Pros and Cons of LLMs. Terry Ewell
presents positive and negative aspects of Large Language Models. An
article from Beijing University is discussed and applied to
college-level education. BDP 372. www.terryewell.com. 2026 Jan.
02.
<Music:
Hummel’s Bassoon Concerto, Second Movement. Terry B. Ewell
and Peter Amstutz, Live Recording, Oct. 1997, Kent State
University.>
Welcome to this brief introduction to Large Language Models, or LLMs,
for my courses. I'm Terry Ewell.
Large Language Models, or LLMs, are changing every day. As I'm making
this video on January 2, 2026, some aspects of it may have already
changed when you're viewing this video. Although there will be
significant advancements in AI and LLMs, I can still draw some
important generalizations. These are based upon my four semesters of
teaching courses at Towson University. My students are required to use
AIN and LLMs in my online and classroom courses. For over five months,
I have intensively worked with ChatGPT. I've employed ChatGPT for my
academic writing, as a fitness and health coach, for financial
tracking, to record my spiritual activities, and for technical
translations into Spanish. I've read academic research and recent
peer-reviewed publications that give me further insights into AI and
LLMs. And last of all, I've gleaned several observations from YouTube
videos and other sources.
This study by Yizhou Fan, conducted in Beijing, China, highlights the
important benefits and disadvantages of using LLMs. Three dimensions of
learning were studied: one, those with human mentors, two, those guided
by checklists, and three, those using LLMs. Participants in this study
used English, which was their second language. These observations were
provided in the study. “ChatGPT's ability to effectively
improve learners' writing performance and productivity had been proven
in many studies.” “ChatGPT improved writing
performance even more than the condition that involved support provided
by a very experienced human expert.” I've seen this in my
courses with written components. The improvement in writing is most
dramatic with students who have weaker language skills. So, in effect,
this is a great equalizer in the course.
I have observed these improved written documents to include better
spelling, grammar, and sentence formation. The citations are in better
form. And overall, the content of the papers is more sophisticated.
Well, let's read on in the study. “Learners with ChatGPT had
significantly higher post-test scores (knowledge gained in the field of
electronic magnetism) than those with human tutors.” A little
further down, “these different studies noted the potential of
AI-powered chatbots, such as ChatGPT, to improve learners' test scores.
knowledge gain and knowledge transfer.”
I've also observed improved test results in my courses. There are even
more benefits than this. There are better research results that I've
found as people have used these. Much better than even searching
online. A more comprehensive presentation of the dilemmas and scenarios
has been a result of using these LLMs. It's easier to find information
with the LLMs. There's an ease of research.
LLMs also handle data very well. For instance, in the medical research
area, the pace of the doubling of data and everything just happened so
quickly that no human could possibly keep abreast of all of the test
results. and LLMs can summarize it very quickly and provide terrific
summaries. So, it's a great time saver. It's a great summarizer of
information.
I'm quite excited about how my students can rapidly improve their
results in my course and also throughout probably the rest of their
lives. Remember, I teach life skills and academic competencies, and as
a result, I must teach AI and LLMs in my courses because these are
important skills that students will need to carry beyond the course
that I teach at Towson University. Employers will be requiring these
skills as well in the future. In all these aspects, LLMs serve as a
lever, an augmentation of what people can do.
There are, however, disadvantages. Having provided the positive aspects
of LLMs, now let's explore the problems as well. First are issues with
hallucinations or errors. The LLMs will produce false data, wrong
facts, and even quote from articles and books that don't exist.
It is important to realize that LLMs are trained to produce human-like
responses, not necessarily ones that are truthful. There's a kind of
eagerness to please by providing a plausible response rather than a
correct reply.
Next is the issue of cognitive offloading. This means a deterioration
of reasoning skills in the users. So let's read further in this study
here. “When learners encounter situations that challenge
their intuition, they are more likely to engage in deliberate
analytical thinking.” “In the context of Gen. AI,
if learners rely excessively on AI-generated outputs or facilitation,
they may not experience the necessary dysfluency or cognitive
difficulty to trigger these deeper metacognitive processes.”
So, in other words, it's talking about struggling to understand
concepts, struggling to figure something out creates in your brain new
neural networks, ways of thinking about things, and processes that are
lost if that cognitive challenge is loaded onto ChatGPT, for instance.
Then you don't have to struggle with it. So, the danger is that the use
of LLMs could lead to surface learning rather than deep learning. There
could be an erosion in reading and research skills. Furthermore,
students may have greater difficulty assessing the validity of
statements and arguments after extensive use of LLMs.
Well, additional problems that emerge with LLMs are plagiarism, which
means passing off another's work, including AI as your own. There is
also the issue of homogenization of responses and language. If everyone
is using LLMs, the results are increasingly similar. There is a loss of
individuality and unique expression. Problems with confidentiality
emerge as well. Those who share private information with LLMs risk that
information being publicized and misused by the company that owns the
LLM. There is, furthermore, the problem of humanization, which is
emotionally connecting to AI as if it were a person. This is a serious
risk for all of us, in particular for those who lack strong and healthy
bonds with other people. All of these are very real and serious
problems. We need to be constantly vigilant to address these issues.
Now, please see my website, which has useful information on AI and
LLMs.
In conclusion, in my courses, we will dive right into the chaotic mix
of the good and the bad, the beautiful and the ugly aspects of AI and
LLMs. In my courses, I will call upon you to actively engage in our
discussions and in our discoveries of how best to use AI and LLMs. The
next videos will present advice on leveraging the benefits of these
while reducing the harmful effects. Thank you.
<Music:
Hummel’s Bassoon Concerto, Second Movement. Terry B. Ewell
and Peter Amstutz, Live Recording, Oct. 1997, Kent State
University.>