I’ve been talking a lot to my computer lately since I installed new operating systems on my machines. I am quite impressed by the dictation accuracy and the degree to which I can use voice commands to control them.
Over the past 40 years I have followed with interest developments in “communication” between humans and computers. In the 60’s I interacted with Eliza, the Rogerian therapist, and in the 70’s the Talking Moose resided on my early Macs—useful toys. But the capabilities of the software to “read” text, translate simple conversations, and follow voice commands have dramatically improved since then and become useful in my work. What was once fiction (e.g. The Circle, 2312, Lexicon) is much closer to (dystopian) reality. The challenge remains how to let technology be a tool controlled by (rather than controlling) me. It is easy to be seduced by the WOW factor.
It’s hard to ignore the stories about ChatGPT that I read about on LinkedIn, the Atlantic, and the New York Times and hear about on NPR. If I were to return to the classroom I would want to explore the strengths and weaknesses of that tool and others like it such as DALL-E.
I recently asked ChatGPT to write a 750-word blog explaining its strengths and weaknesses. Here is what it told me:
ChatGPT is a state-of-the-art language generation model developed by OpenAI. It’s a type of artificial intelligence (AI) model trained on a vast amount of text data to generate human-like responses. This means that when given a prompt, ChatGPT can generate text that follows the same style and tone as the input text.
- Text Generation: ChatGPT’s main function is to generate text based on the input it receives. This could range from generating a response to a simple question to generating an entire article on a given topic.
- Conversational AI: ChatGPT can be used to build conversational AI systems, such as chatbots. These systems can handle tasks such as customer service, information retrieval, and even small talk.
- Question Answering: ChatGPT has been trained on a large corpus of text data, including answers to various questions. This means that it can provide accurate answers to a wide range of questions.
- Summarization: ChatGPT can be used to summarize long pieces of text into a shorter, more concise form. This can be useful for summarizing news articles, scientific papers, or any other lengthy text.
- Translation: ChatGPT can be used to translate text from one language to another. It can be trained in specific languages, and then used to translate text between those languages.
- Text Completion: ChatGPT can be used to complete a partially written text. For example, it can be given the first few sentences of an article, and then generate the rest of the article based on that input.
- Text Classification: ChatGPT can be used to classify text into different categories, such as positive or negative sentiment, topics, and more.
- Text Generation for Specific Applications: ChatGPT can be fine-tuned for specific applications, such as generating creative writing or legal documents.
- Contextual Understanding: While ChatGPT can generate text that appears human-like, it lacks a deep understanding of the context in which the text is generated. This means that it may generate text that is not appropriate for the given context.
- Bias: Like all AI models, ChatGPT is trained on a large corpus of text data, which may include biases. This means that the generated text may reflect these biases.
- Factual Accuracy: ChatGPT is trained on text data, which may not always be accurate. This means that the generated text may contain inaccuracies or false information.
- Lack of Creativity: While ChatGPT can generate text that appears human-like, it cannot truly be creative. This means that the generated text may be repetitive or lack originality.
- Computational Requirements: ChatGPT is a large and complex model requiring significant computational resources to run. This means that it may not be feasible to run on smaller devices, such as smartphones.
In conclusion, ChatGPT is a powerful language generation model that can be used for a wide range of applications. It can generate text, converse with users, answer questions, summarize text, translate text, complete text, and classify text. However, it is important to understand its limitations, including its lack of contextual understanding, potential biases, factual inaccuracies, lack of creativity, and computational requirements.
- OpenAI. (2021). OpenAI API. https://beta.openai.com/docs/api-reference
Here are some resources for Learning More about Chat-GPT that I have found helpful. New articles, new features, and new competing AI software appear daily.
- Apprentice Bard https://www.cnbc.com/2023/01/31/google-testing-chatgpt-like-chatbot-apprentice-bard-with-employees.html?__source=sharebar|linkedin&par=sharebar
- Speaking vs Thinking. https://www.theatlantic.com/technology/archive/2023/01/chatgpt-ai-language-human-computer-grammar-logic/672902/?utm_source=copy-link&utm_medium=social&utm_campaign=share
- DALL-E-2 image creation and art from descriptive natural language: https://openai.com/dall-e-2/
Here are some playful explorations with DALL-E-2