What is ChatGPT and how does it work?
ChatGPT is the latest AI-powered tool to use the power of GPT, the natural language AI model created by OpenAI. But what exactly is it?
AI-powered tools are everywhere, from image retouching to art creation. AI can also have conversations, with ChatGPT being one of the latest chatbots.
The most important thing about ChatGPT is that a human didn’t script it. There are no pre-written answers, databases, or preset tools.
The chatbot can hold conversations, look up facts, review your code (or write it), and attempt to answer anything you ask.
The question is, should we be worried about teaching computers to think in this way?
What is ChatGPT?
The natural language model tends to be on the chatty side. The chatbot could have said things more succinctly.
That said, the response did answer our question. It kept everything in the active voice, a struggle for many writers.
If you ask ChatGPT what it is, you get a pretty comprehensive answer:
ChatGPT is a large language model trained by OpenAI to generate human-like text. It is designed to be able to hold conversations with people, using its knowledge of language and natural communication skills to provide relevant and engaging responses.
With ChatGPT, users can ask questions and have a conversation with a virtual assistant, learning more about a variety of topics and getting answers to their questions in real-time. Whether you’re looking for information, entertainment, or just someone to talk to, ChatGPT is ready and able to assist.
Think of ChatGPT as a general-use Alexa, Siri, or Google Assistant. It understands conversational English, humor, essay structure, and more.
It’s also insanely popular, with over a million people logging into the free service in the first week it was open.
And don’t forget, the more we talk to it, the better it gets as it learns from our questions.
How does ChatGPT work?
If you’ve ever typed a message to a chatbot on a company website, you already know how to use ChatGPT. Those chatbots are scripted or monitored by real humans, but ChatGPT is not.
The AI model can “answer followup questions, admit its mistakes, challenge incorrect premises, and reject inappropriate requests.” The questions can be of almost any form, topic, or task.
OpenAI used books, websites, and articles to train the AI behind the chatbot. Those must have included various subjects, as you can get the AI to do your coding work for you.
The answers you get are generated from probabilities based on your input questions. You can trip it up by asking recent history questions, however, as the dataset only goes up to 2021.
What can you use ChatGPT for?
Honestly, the better question is – what can’t you use it for? ChatGPT can write stories, songs, code, and more.
People have been using it to generate AI art prompts to feed into Midjourney and DALL-E. That gives better-quality images, as the prompts generated are oddly specific.
It might even be better than human writing prompts. After all, the same AI training models that power the majority of AI-generated art are the same models powering ChatGPT.
Another impressive use we noticed was creating a virtual machine inside ChatGPT. That’s right, inside the chat prompt, you can create another computer.
That means you can create a fully-featured computer with an internet connection, a web browser, and more.
When it comes to coding or science questions, the results are even better than the number one search engine, Google Search.
What about Bing? Well, as it turns out, Microsoft has plans to infuse the AI-writing chatbot directly into its search engine.
Do we have to worry about ChatGPT?
Maybe not. While we shouldn’t trust ChatGPT’s response to its potential danger, we can trust other standardized tests.
Take a standard IQ test on which ChatGPT gained a “low average” score of 83. Hardly a super-villain genius level, although it has only been ‘alive’ for a few months, so who knows once it gets more training data.
We’re not sure AI will put people out of work just yet. The number of times it returns incorrect data is still too high. In all cases, you already need an idea of the answer to be able to evaluate ChatGPT’s output.
The AI is only as good as the data sets it is trained on. Those include inherent bias, as does the world it came from.
It could change how work is done. Imagine writing code, then using the AI to debug it. Or the converse, with the AI writing it and you debugging to ensure no errors.
Have any thoughts on this? Carry the discussion over to our Twitter or Facebook.
- When computers meet clutter: Redesigning my office with AI
- White Castle is hiring 100 robots to flip burgers
- This AI lets you talk with your loved ones after they die
- New San Francisco ruling lets police use killer robots (updated)