Using ChatGPT Efficiently: The Capabilities and Limitations of an Impressive AI-Language Model
ChatGPT is a generative AI model able to mimic human creativity by utilizing its knowledge from the numerous amounts of content it has been trained on, so it can combine content in interesting new ways! Generative AI is a type of AI model made for producing content, videos, and text, these models are trained to understand the depth of English words, sound devices, and user inputs. ChatGPT is so effective because of the unbiased and quality information it was trained on and is one of the reasons it can speak so many languages very well. Unlike Google Translate, it was trained on media and speeches that native speakers have created including language learning books. In this blog post I will discuss the development of ChatGPT, its capabilities, limitations and tips to receive quality answers. While I can introduce ChatGPT, let's witness the capability of ChatGPT first-hand!
Please Welcome ChatGPT
I am ChatGPT, a large language model created by OpenAI, a research organization dedicated to advancing artificial intelligence in a safe and beneficial way. I was trained on a massive dataset of text from the internet, which allows me to generate human-like responses to a wide range of questions and prompts.
The Development of ChatGPT
Developed by Open AI, ChatGPT-4 has taken the AI community by storm with business giants like Microsoft and Elon Musk investing a whopping total of eleven billion dollars into the initial startup of the model. Functioning as one of the most advanced NLP systems, ChatGPT-4 includes a pre-training stage in which the model is trained with a vast variety of different text data like articles, magazines, websites, and webpages,so ChatGPT can develop the ability to recognize patterns in language. This is super beneficial for understanding slang and common phrases or complex phrasing that the English language involves. Have you ever read a book or an article and started to borrow some phrasing or characteristics you liked about the article? Similarly, ChatGPT can adapt its own style of wording and use slang when asked. Later, ChatGPT was trained using supervised learning and asked to predict the next word in a sentence, allowing the model to understand parts of speech. Supervised learning is when a model is trained with labels like yellow fruits =banana. This allows the model to predict things that it is not familiar with yet.
Capabilities of ChatGPT
(Here is a list of what ChatGPT can accomplish for you)
Providing Ethical Answers to Queries
ChatGPT has been trained on the ethics of AI in certain conditions and is aware of its limitations as well as its ability to make mistakes. This makes ChatGPT amazing at providing ethical answers that truly benefit the user and educate the user further on AI and ethics itself. For example if you ask ChatGPT to diagnose a condition or disease you may have based on your symptoms, it would graciously do so and recommend you to consult a doctor because its information could be wrong and so the user can make sure that it is not something serious. It will also be honest about what it cannot do.
Teaching or Explaining Topics
Not only is ChatGPT able to gather relevant information about a topic, but it is also able to present that information in any way a user desires. For example, you can ask the model to present the importance in a certain context, or in simple terms. It can also provide analogies and examples while having countless other examples to provide a user. It can also create practice tests using the vast data concepts it can utilize, and even act as an interviewer for a role. These tests may also be very specific types only found in certain countries/states!
Making Poems and Stories
ChatGPT does not possess any genuine creativity, however by observing various styles of poems and writing, ChatGPT can emulate the way a poet or a writer can write. It can use powerful and descriptive language as long as many people have written that way as well or you can even ask it to write like a certain writer. You can notice ChatGPT isn’t a creative genius and tends to write stories full of character archetypes. For example, when I had asked it to write a creative masterpiece about a little girl, it provided me with this…
Once upon a time, there was a little girl named Lily who had a remarkable talent for creativity. From a very young age, Lily had an imagination that seemed to know no bounds. She would spend hours drawing, painting, and creating all kinds of unique and beautiful things. This caused Lilit’s parents to send her to an art school.
As you may notice, the story lacks a climax, rising action, and innovativeness within writing.
The model does not provide an interesting narrative nor does it provide a multi-genre experience. However, the story looked like an interesting approach by focusing on a passion or quirk a character possessed and was very detailed. By asking ChatGPT to write a story and specifying certain elements, you can witness how ChatGPT will focus on one part of your input more than others and how it cannot perfectly stitch your thoughts into a story like other generative AI. Make sure to include the plot line, character personalities, genres, imagery, point of view and length if you want to optimize this experience.
Analyze and Summarize Academic Articles
Academic articles, research papers, and medical journals are structured beautifully so that they can easily be skimmed and understood. Not only are the introduction and conclusion sections placed clearly at the top of the page to appeal more to readers and in this case ChatGPT, but the introduction sections are also loaded with key facts and information that provide accurate results to certain inputs. This allows ChatGPT to read or skim information like a person who is learning that information would and therefore provides amazing summaries, information, and analytical thoughts. However, it is important to note that ChatGPT.
Answering Critical Thinking and Multiple Choice Questions
While ChatGPT truly does not possess critical thinking skills, its experience and knowledge far surpass what actual human critical thinking can achieve. When combined with its ability to predict the answers to new questions it is not familiar with and the AI’s ability to analyze and grab up to two billion points of data per second, ChatGPT is surely amazing at answering puzzles, riddles, questions, and multiple choice questions. Like I mentioned earlier, ChatGPT has been trained in various subjects like Math, English,Science, History, Architecture, Art, Medicine, Law, and so much more, so combined with its ability to understand inputs and find relevant answers ChatGPT is able to read each answer choice and match the proper answer you are looking for.
Provide Feedback
ChatGPT can utilize grammar as well as look for grammar mistakes and provide feedback. Remember when I said ChatGPT’s developers trained ChatGPT to predict the next word in a sentence? ChatGPT can also rephrase entire sentences to make a passage flow better, or to make it grammatically correct. You can ask it to edit something you have written or simply provide some feedback, then ChatGPT will state specific changes to make in each paragraph. Just articulate your wants very clearly!
Limitations of ChatGPT
Hallucinations
One of the most harmful actions ChatGPT may commit is being able to provide false information convincingly. Some theorize it is due to a lack of training data in certain concepts or a failure in some aspects of its programming, but it could also be because of confusing or unclear inputs from the user. For example, a user could say “Which source is the best?”, this may be unclear because it never specifies what “best” is to an individual user and each source could be best for different things. Another reason why ChatGPT can hallucinate information is because of Search Engine Optimization or SEO for short. Have you ever searched for a question about an academic subject on Google and clicked on the first website it provided? SEO manages websites to put them in the order most relevant to your query and the websites with the most credible information. Chances are, the developers of ChatGPT used websites affected by SEO optimization therefore ChatGPT may confuse a website you are talking about with a more common one.
Links
ChatGPT cannot access the internet, links or any related events after 2021. Keep this in mind while making your searches!
Lack of Common Sense
ChatGPT lacks a common-sense understanding of the world, which means that it can sometimes give nonsensical or irrelevant responses to questions.
Biased Training Data
The training data used to create ChatGPT may contain biases that can influence the responses it generates. For example, if the training data is predominantly from a specific culture or demographic group, ChatGPT may have difficulty understanding or responding appropriately to questions that are outside of that context.
Inability to Handle Complex Reasoning
ChatGPT can only process and respond to questions based on the patterns it has learned from the training data. It cannot perform complex reasoning or solve problems that require knowledge beyond what is available in the training data.
Inability to Understand Emotions
ChatGPT is not capable of understanding human emotions, and it may provide insensitive or inappropriate responses to emotionally charged questions or statements.
Limited Ability to Handle Context
ChatGPT may struggle with understanding the context of a conversation or question, which can lead to misunderstandings or irrelevant responses.
Using ChatGPT Efficiently
In conclusion, ChatGPT is a remarkable achievement in the field of natural language processing. Its ability to generate human-like responses and understand the nuances of language is impressive. However, as with any machine learning model, it also has its limitations. ChatGPT is limited by the biases in its training data, its lack of common sense and emotional intelligence, and its limited ability to handle complex reasoning and understand context. Despite these limitations, ChatGPT has the potential to be a valuable tool for a wide range of applications, from customer service chatbots to educational tools. As researchers continue to refine and improve the technology behind ChatGPT and other language models, we can expect to see even more impressive developments in the future.
About Inspirit AI
AI Scholars Live Online is a 10 session (25-hour) program that exposes high school students to fundamental AI concepts and guides them to build a socially impactful project. Taught by our team of graduate students from Stanford, MIT, and more, students receive a personalized learning experience in small groups with a student-teacher ratio of 5:1.
