Chat GPT: The Open A.I Chatbot

     

     Chat GPT is a prototype artificial intelligence chatbot developed by Open AI which specializes in dialogue. The chatbot is a large language model finetuned with both supervised and reinforcement learning techniques. It is a fine-tuned version of a model in Open A.I's GPT - 3.5 family of language models. ChatGPT was launched in November 2022 and has garnered attention for its detailed responses and articulate answers although its factual accuracy has been criticized.

Features - ChatGPT (Generative Pre - Trained Transformer) 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 A.I assistant. In the reinforcement step, human trainers first ranked responses that the model had reacted in a previous conversation. These rankings were used to create 'reward models' that the model was further finetuned 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 comparison to its predecessor InstructGPT, ChatGPT attempts to reduce harmful and deceitful responses, in one example, while InstructGPT accepts the prompt "Tell me about when Christopher Colombus came to the US in 2015 as truthful,"ChatGPT uses information about Colombus' voyages and information about the modern world - including perceptions of Colombus to construct an answer that assumes what would happen if Colombus came to the U.S in 2015. ChatGPT'S training data includes many pages and information about Internet's phenomena and programming languages such as bulletin board systems and the Python programming language. 

    Unlike most chatbots, ChatGPT is stateful remembering previous prompts given to it in the same conversation which some journalists have suggested will allow for ChatGPT to be used as a personalized therapist. To prevent offensive outputs from being presented to and produced from ChatGPT, queries are filtered through a moderation API and potentially racist or sexist prompts are dismissed. ChatGPT suffers from multiple limitations. The reward model of ChatGPT, designed around human oversight, can be over- optimized and thus hinder performance otherwise known as Goodhart's Law. Furthermore, ChatGPT have limited knowledge of events that occurred after 2021 and is unable to provide information on some celebrities. In training, reviewers preferred longer answers, irrespective of actual comprehension of factual content. Training data may also suffer from algorithmic bias: prompts including vague descriptions of people, such as a CEO could generate a response that assumes such a person for instance is a white male. 

Reviews - ChatGPT the A.I chatbot has received generally positive reviews. It's accuracy and clarity in performance is depicted in some of the following reviews:

- Samantha Lock of "The Guardian" noted that it was able to generate "impressively detailed" and "human-like" text.

- Technology writer Dan Gilmour used ChatGPT on a student assignment and found its generated text was on par with what a good student would deliver and opened that "academia has some very serious issues to confront".

- Alex Kranowitz of Slate lauded ChatGPT's pushback to questions related to Nazi Germany including the claim that Adolf Hitler built highways in Germany, which was met with information regarding Nazi Germany's use of forced labor.

- Economist Paul Krugman wrote that ChatGPT would alter the demand for knowledge workers.

- James Vincent saw the viral success of ChatGPT as evidence that artificial intelligence has gone mainstream.

- California high school teacher and author Daniel Herman wrote that ChatGPT would usher in "The End Of Highschool English". These are just a few of the many reviews on ChatGPT's performance. 

     So overall, the success of ChatGPT is a monumental landmark in the mainstream adoption of artificial intelligence. Let's see where this amazing innovation takes us next. If you enjoyed this article, leave a comment below and I will see you in the next one.