ChatGPT, an artificial intelligence (AI)-based chatbot tool from OpenAI, is one of the hottest topics on Twitter social media. ChatGPT is widely discussed because this chatbot with artificial intelligence can answer various questions flexibly. Flexible here means that ChatGPT can understand the context of the conversation and provide response text according to context, so it's not rigid like a robot.
One of them is the Twitter account that discusses ChatGPT from OpenAI, @huedaya. He created a thread about what ChatGPT is with some testing is done. One of the experiments was to ask several questions and ask to correct English grammar or grammar in a list format. "ChatGPT also managed to find several errors in my writing. Complete with error reasons," tweeted @huedaya in one of his threads, when he "challenged" ChatGPT to correct English grammar.
Also read: Artificial Intelligence Can Detect Covid-19 Through Cough Sounds Some Twitter users compare ChatGPT results with Google search results. "Google is finished! Try to compare the response quality (between Google and ChatGPT)" tweeted one Twitter user with the account @jdkelly.
Some Twitter users also compared ChatGPT results with Google search results. "Google is finished! Try to compare the response quality (between Google and ChatGPT)" tweeted one Twitter user with the account @jdkelly.
What is ChatGPT? ChatGPT is an artificial intelligence (AI) language model developed by OpenAI, a company engaged in the development of AI technology.
ChatGPT is a chatbot based on GPT-3.5, a natural language model that uses deep learning processes. Natural here means the text style generated by the chat robot made by OpenAI, which is similar to the writing style of humans naturally, not rigid like machines or bots. So naturally, ChatGPT can also be asked to compose poetry.
Like chatbots in general, ChatGPT made by OpenAI allows users to send questions or instructions which ChatGPT will respond to later. Currently, ChatGPT also supports Indonesian, so users can submit and receive responses in Indonesian.
Interestingly, ChatGPT from OpenAI is designed to remember questions that have been asked before and can correct themselves according to the feedback provided by the user. However, the questions asked should not be over the line. This is because ChatGPT is also trained to reduce inappropriate questions from users.
So far, OpenAI has only opened the bot for beta testing and evaluation. Meanwhile, new API access will be available next year. With API access, developers can implement ChatGPT into their software, as KompasTekno summarized from Indian Express, Monday (5/12/2022).
ChatGPT is not perfect
Although many people are amazed by the results of ChatGPT, however, this chatbot is also not perfect. ChatGPT is still prone to misinformation and bias. OpenAI understands this shortcoming and listed it on the blog.
"ChatGPT can write answers that sound reasonable, but (can) be wrong or don't make sense," wrote OpenAI on its blog. OpenAI said, there are several challenges to solving this problem. First, during the Reinforcement Learning (RL) process, there is currently no source of truth. To note, ChatGPT uses the Reinforcement Learning from Human Feedback (RLHF) method, or a machine learning model that allows the software to study its environment according to the feedback provided by the user.
The second challenge is to train the algorithm carefully, otherwise, it can make the machine reject questions that can be answered correctly. The final challenge according to OpenAI is that the ideal answer according to the bot depends on what the algorithm has learned, not what humans know. If you are curious to try ChatGPT, you can access the following link and continue creating an account first. After successfully logging in, select the Playground menu to try giving questions or instructions.