What do we mean by a large language model? In this blog post, I will define my understanding of it after reading several books and articles about Artificial Intelligence. I will also explore the benefits and concerns regarding large language models and how we might mitigate some of those concerns.
Many of us, including myself, became aware of the Large Language model in October 2022, when Open AI made chat-GPT available for public use, which I, like many others, did. I will link to a post I wrote at the time based on this experience.
Many large language models are available, including chat-GPT and Google Gemini.
What is a Large Language Model?
A large language model uses statistical analysis to predict the next word to generate text. It is trained using a combination of Machine Learning and Human input. Machine learning is an offshoot of Artificial Intelligence, which Algorithms can learn from feedback.
The Large Language model is built on an underlying neural network called a Transformer. A neural network is like the human brain, comprised of interconnected neurons. But not yet to the same level of complexity as the human minds
One of the biggest downsides of the large language model is its capability to supply misinformation. However, this risk can be reduced by asking the model to explain its answers. We will explore this further.
The benefits of using a Large Language Model
Why don’t we consider the potential benefits of using large language models, both for us individually and for our wider society?
One potential benefit is that Artificial Intelligence tools like large language models will lower the barriers to learning, helping more people fulfil their potential abilities. For example, a large language model could help children organize their work. At the latter stage, if it’s a written task, the large language model could provide feedback on the student’s work, especially on their grammar.
You can use them individually to help you lay out a piece of content, brainstorm ideas, and summarise your notes, while others use them to help with research.
Programmers are using them to write initial code and to identify bugs in pre-existing code.
The problems of using a Large Language Model
The most obvious problem is that you can’t trust any answer it provides, as it could be full of misinformation. If you don’t fact-check, you could be left to feel pretty stupid. This problem can be reduced by asking the Large Language model to provide its reasoning.
As the data used in training a large language model is human data, it learns from the biases in our society, some of which may be historical, and uses this bias in our answers. Yes, it is a problem, but not with the large language model itself but with the society we live in.
The fact that a large language model is so good at English might prevent people from writing. One of the many benefits of writing is that it forces the writer to decompress their thoughts into ideas that can be shared in an abstract form of written language. This process forces us to think about what we are writing, and occasionally, we will get a fresh insight.
For example, I wrote a blog post about information overload. During the writing process, I gained an insight that completely changed my thoughts. Before I wrote that post, I saw information overload as a threat to the information golden age we live in. I now realise that it is a side effect of our information golden age that we must learn to manage.
I may never have gotten that insight if I had never written that blog post. That is why I worry that the Large Language Model will stop us from writing.
The underlying large language model is seated in a Black Box. No one really understands how a large language model works, and because of this, it tends to surprise both us and the companies that develop it.
The process of unexpected abilities emerging is known as emergence. Examples of known abilities that emerged from large language models include:
- Multiplication
- Generating computer code in a range of different programming languages
- Simulating a Linux terminal
The benefits of using Large Language models
One of the benefits of lowering barriers to learning is that a large language model can help you improve your writing skills. You could use it to assess your own writing and request suggestions on how it can be improved, especially if English isn’t your first language.
Or, if you are studying a complex piece of text that you are struggling to understand, you could ask the AI to summarise the text, which could help you break down and define your own reasoning from the original.
Another benefit is the potential to use large language models in personalised education tools. Some note-taking tools, including Notion and Google Notebook LM, have implemented large language models.
A case study: how I use Artificial Intelligence
I have been playing around with a number of large language models, either via the model’s own chatbot or embedded into other applications. I will include a link to the Artificial Intelligence section of the CTNET website, where all our Artificial Intelligence articles are located.
During that time, I experimented with using large language models to help me with my blogging by asking them to lay out the possible structure of the blog posts. I would also use AI to write my Google metadata and social media posts to promote my new content.
It worked reasonably well, with me editing the responses most of the time and occasionally using what the AI had produced.
I have used the Smart Connections Obsidian community plugin for the last few months. It does all this by allowing me to combine the notes in my Personal Knowledge Management system with a large language model such as GPT-4o, which I’m currently using.
The layout of blog posts can be a bit hit-and-miss, but if it’s a subject that I have a lot of notes about, it works really well. I have included things in blog posts that I don’t think I would have done without it prompting me to. In fact, this very section of this blog post wouldn’t have been deleted without the AI suggesting that I share my own experiences as a community plug-in.
Obsidian Smart Connections is also useful for linking the notes in my Zettelkasten to related thoughts and ideas.
As I stated in my post about my concerns about using Artificial Intelligence in a Personal Knowledge Management System, you will allow it to form the links, reducing the impact of finding those relationships and the potential for your mind to form its own connections.
https://www.ctnet.co.uk/obsidian-smart-connections-ai-plugin/I have tried to adhere to this, but on occasion, I see a suggested link to a note I had forgotten, and on occasion, I find two notes that are basically around the same idea, allowing me to link them together. You can find out more about the Smart Connections community plug-in guide.
Conclusion
The Large Language Model is a technological tool that is still being actively developed. It will greatly impact how we learn, work, and play.
We need to learn about the negative and positive impacts technology will have on us, as language is one of the key things that makes us human and how we see the world around us.
Use the Large Language Model wisely and with the intention to help you grow and learn.
I hope you have enjoyed this post. I would appreciate it if you could fill out this form and help us improve our future content.