ChatGPT is a chatbot launched by OpenAI in November 2022. It is built on top of OpenAI's GPT-3 family of large language models, and is fine-tuned with both supervised and reinforcement learning techniques.
What is Chat GPT?
Chat GPT is defined as a generative language model. However in practice it is understood as an artificial intelligence chat that has been trained and designed to hold natural conversations. Chat GPT belongs to the research company OpenAI, founded in San Francisco in 2015 by Sam Altman, Elon Musk, Greg Brockman, Ilya Sutskever and Wojciech Zaremba.
How does Chat GPT work?
As its acronym indicates, Generative Pre-training Transformer, Chat GPT is a generative language model based on the ‘transformer’ architecture. These models are capable of processing large amounts of text and learning to perform natural language processing tasks very effectively. The GPT-3 model, in particular, is 175 billion parameters in size, making it the largest language model ever trained. To work, GPT needs to be “trained” on a large amount of text. For example, the GPT-3 model was trained on a text set that included over 8 million documents and over 10 billion words. From this text, the model learns to perform natural language processing tasks and generate coherent, well-written text. Once the model is well trained, GPT can be used to perform a wide range of tasks, as we have seen in the previous section. Reinforcement learning, based on human feedback, was used for training. Ultimately, by supervised fine tuning. The human AI trainers provided conversations in which they represented both the user and the AI assistant. In addition, the coaches were provided with written suggestions to help them write their proposals. So, they mixed this new dataset with the InstructGPT dataset that was transformed into a dialog format.
But how did they create the reward model for reinforcement learning?
The first thing that was needed was to collect comparison data. This consisted of two or more model responses, ranked by quality. So, in order to collect the data, they took some conversations that the trainers had had with Chat GPT and randomly selected them. In this way they tested various endings for the coaches to rank.
For this reason, these reward models could be adjusted using Proximal Policy Optimization. Also, the trainings were carried out on a Microsoft Azure platform on a supercomputer. In conclusion, to use GPT in a chat, the model is provided with an input in the form of text. This input can be in the form of a question or a context sentence. And, from this input, GPT generates an appropriate and coherent response. In fact, this response can be used in a chatbot or any other application where it is necessary to generate a text from a given input.
Is ChatGPT free?
ChatGPT is the latest iteration of GPT (Generative Pre-Trained Transformer), a family of text-generating AI programs. It's currently free to use as a “research preview” on OpenAI's website but the company wants to find ways to monetize the tool
Disclaimer:
***The following video abides by the YouTube Community Guideline. Footage used in this video is for educational purposes.
Portions of stock footage of products were gathered from multiple sources including, manufactures, fellow creators and various other sources.
COPYRIGHT ISSUE: If you can find any copyright infringement then send us an email. All rights reserved by respective owners.
***The footage used in this video follows Fair Usage Policy “Under Section 107” of the “Copyright Act 1976”. If you have any copyright issues, please send us an email or let us know by commenting below.