Features
The new version makes it possible for ChatGPT to respond to follow-up inquiries and accept mistakes. It can also refute false assumptions about anything and reject improper requests thanks to its efficient dialogue style. It can also write poems and fix coding issues with thorough explanations. It can come up with AI art prompts, write fresh code, and explain the complexities of the algorithms with ease.
A lot of potentially harmful queries, for example, things like hate speech or violent and antisocial content, are left unanswered because doing so would go against its programming and purpose. This will further allow all the users of this interface to interact with the AI model without fearing any harm caused or any information leaked.
Challenges
One of the main issues limiting large language models’ potential commercial utility is how to control the aggressive tendencies of internet users. OpenAI, with ChatGPT, aims to put forward a substantial incremental step toward creating a safe AI model for everyone by using machine learning to generate specific content filters against trolling behaviour.
They do this with the help of a method known as Reinforcement Learning from Human Feedback (RLHF). This method adjusts an AI model based on how people respond to its generated responses and goes through an extensive iterative process to make it smarter.
Unfortunately, there is no limit to human ingenuity, even when it comes to bad things, and it seems like trolls and hackers have already learned to bypass such filters by using social or phishing attacks that limit the output of the program to a fake and pretentious scenario or even a fake poem or story.
Availability
Because it is currently in a testing phase, ChatGPT is currently free to use with an OpenAI account. As the days go by, OpenAI will also announce more advanced and paid versions of the AI application.
