Announcement of language model 'ChatGPT' for dialogue, it is possible to admit mistakes and reject inappropriate requests

OpenAI, which developed the high-performance language model ``

OpenAI debuts ChatGPT and GPT-3.5 series as GPT-4 rumors fly | VentureBeat
https://venturebeat.com/ai/openai-debuts-chatgpt-and-gpt-3-5-series-as-gpt-4-rumors-fly/

ChatGPT is a GPT-3 chatbot from OpenAI that you can test now
https://the-decoder.com/chatgpt-is-a-gpt-3-chatbot-from-openai-that-you-can-test-now/

ChatGPT, like InstructGPT , an improved version of GPT-3, is trained using Reinforcement Learning from Human Feedback (RLHF).

Features are interactively optimized, where you can answer follow-up questions, admit mistakes, challenge false assumptions, and reject inappropriate requests. .

However, since it is still in the testing stage, it may show answers that sound plausible but are inaccurate or meaningless.

In addition, since it is sensitive to subtle changes in the input content, if you cannot get an appropriate answer to a question, you may get the correct answer by changing the wording or paraphrasing.

In addition, although inappropriate requests are rejected, it may respond to harmful instructions or behave biasedly.

Anyone with an OpenAI ID can participate in the test for free during the preview period.

ChatGPT
https://chat.openai.com/