| ChatGPT For Dummies |
Page 1 of 4 Author: Pam Baker
There has been a massive amount of interest in ChatGPT since its release. This short book aims to describe what ChatGPT is, how it works, how to get the best from it, how it might change many industries and professions, and how it can improve your life. In many ways ChatGPT acts as a junior assistant, quickly producing useful results, but the output needs to be fact-checked. The book is aimed at beginners, having little or no knowledge of ChatGPT, but having some basic computing knowledge (e.g. browsers, web apps, computing devices). Below is a chapter-by-chapter exploration of the topics covered. 
Chapter 1: Introducing ChatGPT The book opens with an overview of this new technology called Generative AI that can generate new content and analyze content – based on natural human like language queries working on massive amounts of data (Large Language Models). We’re quickly into setting up an OpenAI account. OpenAI is the company behind ChatGPT, and Microsoft has invested heavily in it – hence the fast integration of ChatGPT into Microsoft products. After setting up an account you can start issuing queries (called prompts), and subsequent queries within the same chat session can make use of your previous questions and replies (i.e. it uses context). You can use the thumbs up/down to indicate how satisfied you are with the answers – and these responses feed back into ChatGPT. Note your data may be used for further training. It should be noted that the responses may not be correct! It can be wrong (called hallucination), biased, and stupid at times. That said, the key to getting good answers is to create a good prompt (i.e. asking the right questions). In many ways, it’s like having a junior assistant, the output is often very good, but it needs checking. The interaction with the ChatGPT session seems like conversation. The chapter moves onto what ChatGPT is and isn’t. It isn’t General AI (i.e. human-level intelligence), it largely works by guessing the next response word based on pretraining on massive datasets. It works with probabilities based on patterns to make informed guesses. This pattern recognition supports at least 95 languages, and various programming languages. The author acknowledges there are fears about jobs, but points out the winners are likely to be employees that can use ChatGPT effectively. Various beneficial uses of ChatGPT are outlined (e.g. help diagnose illnesses). It’s suggested it will affect and replace some jobs, but people have the advantage of creativity/intuition (I do wonder if this is true). There is a list of industries that may be affected by ChatGPT – but no explanations as to why are provided. Some interesting uses are proposed, including: interviewing a dead person, reply as a given person or manner (e.g. strict/humourous), recommend colours for logos, generate original work for articles/books, medical diagnosis, summarise issues or books, teach a skill, write a resume/CV, write a speech, plan a party/trip. You can clearly see the potential this tool can offer. There’s a very useful table of the pros/cons of using ChatGPT. The content is generally easy to read and interesting (the author is a journalist), with a good outline of what ChatGPT is, and some of its uses. Lots of information is packed into this initial chapter. The book’s other chapters go into more detail on how ChatGPT works, how to get more out of it, how it may affect certain industries, and its place in the wider world of AI. Chapter 2: Discovering How ChatGPT Works Here we look at what makes ChatGPT different. Unlike search engines, that retrieve existing information, ChatGPT can create new content based on your prompt. Learning how to write the correct prompt is the key to using the tool optimally. ChatGPT creates responses by using context and assigning weights to words, to predict the next likely words. Next, there’s a look at the architecture of ChatGPT. Underlying ChatGPT are Large Language Models (LLM) which are trained on masses of information, and this can be used in creating novel responses. ChatGPT 4 has billions of parameters – where a parameter is a value that has a weight and connection in the underlying neural network architecture. ChatGPT uses multilayer transformers (neural networks), which can efficiently run tasks in parallel aiding performance. An important aspect of transformers is self-attention, which understands various representations of a given word (i.e. it understands context). Training the model is expensive, using massive amounts of resources, one supercomputer mentioned has 285,000 CPUs and 10,000 GPUs. Training is followed by reinforcement learning from human feedback (RLHF) – this is the purpose of the thumbs up/down icons in ChatGPT’s responses, you can provide feedback on how good the answer is. A useful list of ChatGPT’s limitations is provided, including:
ChatGPT is being integrated into lots of existing software. Perhaps to be expected, since it helps fund OpenAI, Microsoft’s Bing and Office 365 products have ChatGPT inbuilt. Various other vender add-ins are briefly discussed. There’s a useful table listing the differences between search engines and ChatGPT. The chapter ends with a look at how new businesses can be built on ChatGPT. In addition to your own business ideas, you can ask ChatGPT are innovative ideas to create profitable businesses (e.g. a new way to generate renewable energy). It can automate business communications (e.g. answering emails). It can be used to generate ideas for a book, create first drafts of books - but remember to fact check everything! While the book says ChatGPT doesn’t offer references or sources for its responses, this is no longer true, you can ask it to specify its sources for its answers, and it's included automatically in Bing. |
|||||
| Last Updated ( Tuesday, 24 October 2023 ) |