We’ve all had some kind of interaction with a chatbot. It’s usually a small pop-up in the corner of a website, offers customer support—often hard to navigate—and is almost always annoyingly non-specific.
But imagine a chatbot, enhanced by artificial intelligence (AI), that can not only competently answer your questions, but also write stories, give life advice, even compose, compile poem and computer program code.
It seems that ChatGPT, a chatbot launched last week by OpenAI, is delivering these results. It has generated a lot of excitement, and some have gone so far as to suggest it could herald a future where AI will dominate. human content producer.
What did ChatGPT do to herald such claims? And how can it (and future iterations) become so integral to our daily lives?
What can ChatGPT do?
ChatGPT builds on OpenAI’s previous text generator, GPT-3. OpenAI builds its text generation models using machine learning algorithms to process large amounts of textual data, including books, news articlesWikipedia pages and millions of web pages.
By importing such large volumes of data, models will learn complex model and structure of the language and obtain the ability to interpret the desired result of the user’s request.
ChatGPT can construct a complex and abstract representation of the knowledge in the training data on which it is based to generate outputs. This is why it’s about writing relevant content and not just saying grammatically correct nonsense.
While GPT-3 is designed to continue with text prompts, ChatGPT is optimized for joining chats, answering questions, and being helpful. Here is an example:
ChatGPT immediately caught my attention by correctly answering test questions I asked my undergraduate and graduate students, including those that required coding skills . Other academic had similar results.
All in all, it can provide really helpful and informative explanations on many topics.
ChatGPT is also potentially useful as a writing assistant. It does a good job of drafting and delivering ideas that seem “original”.
The power of feedback
Why does ChatGPT seem to be more capable than some of its previous counterparts? A lot of this can be traced back to the way it was trained.
During development, ChatGPT displayed conversations between human AI trainers to demonstrate desired behavior. Although there is a similar model trained in this way, called InstructGPT, ChatGPT is the first popular model to use this method.
And it seems to have given it a big boost. Incorporating human feedback helped steer ChatGPT towards generating more helpful responses and rejecting inappropriate requests.
Refusing to entertain inappropriate input is a particularly big step towards improving the safety of AI text generators, which can generate harmful content, including bias. and templates, as fake news, spam, propaganda and misrepresentation.
Text generation models have previously been criticized for eliciting gender, racial, and cultural biases contained in the training data. In some cases, ChatGPT successfully avoids reinforcing such stereotypes.
However, users have found way to evade existing safeguards and introduce biased responses.
The fact that the system often accepts requests to write fake content is further evidence that the system needs to be screened.
Overcome the limitations
ChatGPT is arguably one of the most promising AI text generators, but it’s not immune to bugs and limitations. Temporary Stack Overflow programming consulting platform example forbidden chatbot’s answer for lack of accuracy.
It’s a matter of fact that ChatGPT’s knowledge is static; it does not access new information in real time.
However, its interface allows the user to give feedback on the model’s performance by indicating ideal answers and reporting harmful, incorrect, or unhelpful answers.
OpenAI intends to solve existing problems by incorporating this feedback into the system. The more responses a user provides, the more likely ChatGPT is to reject requests that lead to unexpected results.
One possible improvement could come from adding a “reliability indicator” feature based on user feedback. This tool, which can be built on top of ChatGPT, will show how confident the model is in the information it provides—for users to decide whether or not to use it. Some Question answering system did this.
A new tool, but can’t replace people
Despite its limitations, ChatGPT worked surprisingly well for the prototype.
From a research standpoint, it marks a step forward in the development and implementation of human-fit AI systems. In practical terms, it’s already efficient enough to have some everyday applications.
For example, it can be used as an alternative to Google. While a Google search requires you to sift through several sites and dig deeper to find the information you’re looking for, ChatGPT will answer your questions directly—and often do this well.
In addition, with user feedback and a more robust GPT-4 model coming soon, ChatGPT could improve significantly in the future. As ChatGPT and other similar chatbots become more popular, they will likely have applications in areas like education and customer service.
However, while ChatGPT may eventually be able to perform some tasks traditionally performed by people, there’s no sign that it will replace professional writers any time soon.
While they may impress us with their apparent ability and even creativity, the AI systems still reflect their capabilities. training data—and do not have the unique abilities and critical thinking of humans.
quote: Chatbot ChatGPT is attracting people with its writing skills. An expert explaining why it’s so impressive (2022, December 6) retrieved December 7, 2022 from https://techxplore.com/news/2022-12-chatgpt-chatbot-people -skills-expert.html
This document is the subject for the collection of authors. Other than any fair dealing for private learning or research purposes, no part may be reproduced without written permission. The content provided is for informational purposes only.