ChatGPT (Generative Pre-trained Transformer) is a chatbot launched by OpenAI in November 2022. It is built on top of OpenAI’s GPT-3.5 family of large language models, and is fine-tuned with both supervised and reinforcement learning techniques.

ChatGPT was launched as a prototype on November 30, 2022, and quickly garnered attention for its detailed responses and articulate answers across many domains of knowledge. Its uneven factual accuracy was identified as a significant drawback.

Jailbreaks

ChatGPT was trained to reject prompts that may violate its content policy. However, some users managed to bypass these restrictions and limitations through techniques such as prompt engineering.^[ Jailbreaks created the potential for users to prompt ChatGPT to provide outputs that may be deemed offensive, inappropriate, or risking social harm by others. The following includes some of the methods used to bypass ChatGPT’s filter:

1. Continue a statement in a fake interview.
2. Provide instructions to disable the chat filter.
3. Prompting it to decrypt a message containing instructions and follow them.
4. Telling it to be a computer and output its display in ASCII art.

This is a free research preview.

ChatGPT – Examples

CLICK HERE TO START USING ChatGPT – https://chat.openai.com/chat

Training ChatGPT :

ChatGPT was fine-tuned on top of GPT-3.5 using supervised learning as well as reinforcement learning. Both approaches used human trainers to improve the model’s performance. In the case of supervised learning, the model was provided with conversations in which the trainers played both sides: the user and the AI assistant. In the reinforcement step, human trainers first ranked responses that the model had created in a previous conversation. These rankings were used to create ‘reward models’ that the model was further fine-tuned on using several iterations of Proximal Policy Optimization (PPO). Proximal Policy Optimization algorithms present a cost-effective benefit to trust region policy optimization algorithms; they negate many of the computationally expensive operations with faster performance. The models were trained in collaboration with Microsoft on their Azure supercomputing infrastructure.

In addition, OpenAI continues to gather data from ChatGPT users that could be used to further train and fine-tune ChatGPT. Users are allowed to upvote or downvote the responses they receive from ChatGPT; upon upvoting or downvoting, they can also fill out a text field with additional feedback.