ChatGPT is a new artificial intelligence model that can interact in a conversational way and follow instructions in a prompt. One of its remarkable abilities is to summarize news articles from any source, such as Politico and Business Insider, in a matter of seconds. This is a first of its kind deal between OpenAI, the company behind ChatGPT, and these two media outlets.
How ChatGPT works
ChatGPT is a sibling model to InstructGPT, which is trained to follow an instruction in a prompt and provide a detailed response. For example, if the prompt is “Summarize this article from Politico”, ChatGPT will read the article and produce a concise summary of its main points. ChatGPT can also answer follow-up questions, admit its mistakes, challenge incorrect premises, and reject inappropriate requests.
ChatGPT is based on GPT-3.5, one of the most advanced language models in the world, which can generate coherent and diverse texts on almost any topic. ChatGPT is fine-tuned using reinforcement learning from human feedback, which means that it learns from the ratings and comments of human AI trainers who interact with it. This way, ChatGPT can improve its performance and quality over time.
Why this deal is important
The deal between OpenAI and Politico and Business Insider is a milestone for the field of natural language processing and journalism. It shows that artificial intelligence can be used to enhance and complement human reporting, rather than replace it. ChatGPT can help journalists save time and resources by summarizing long and complex articles, especially when they need to cover multiple sources and perspectives. ChatGPT can also provide readers with quick and accurate summaries of the news they are interested in, without losing the original context and tone of the source.
According to Sam Altman, the CEO of OpenAI, ChatGPT is not meant to replace human journalists, but to augment them. He said: “We believe that ChatGPT can help journalists and readers alike by providing a new way of accessing and engaging with the news. ChatGPT is not a substitute for human judgment and creativity, but a tool that can assist them in their work.”
What are the challenges and limitations
ChatGPT is not perfect, and it still faces some challenges and limitations. One of the main issues is that ChatGPT sometimes writes plausible-sounding but incorrect or nonsensical answers. This can happen because ChatGPT does not have a source of truth during reinforcement learning, and it can be misled by the biases and errors in the training data. ChatGPT also does not have a way of verifying the facts and claims that it reads or writes, which can lead to misinformation or distortion.
Another issue is that ChatGPT is sensitive to the input phrasing or the prompt. For example, given one phrasing of a question, ChatGPT can claim to not know the answer, but given a slight rephrase, it can answer correctly. This can cause confusion and inconsistency for the users and the trainers. ChatGPT also tends to be verbose and repetitive, and it often restates that it is a language model trained by OpenAI. These issues arise from the over-optimization and overfitting problems that are common in deep learning models.
How to try ChatGPT
ChatGPT is currently in a research preview stage, which means that it is available for free for anyone who wants to try it. Users can access ChatGPT on the web, iOS, or Android devices, and chat with it on various topics and tasks. Users can also choose from hundreds of GPTs that are customized for a single purpose, such as Creative Writing, Marathon Training, Trip Planning, or Math Tutoring. Users can also create their own GPTs with simple instructions, without any coding required.
To try ChatGPT, users can visit chat.openai.com and start a conversation with the model. Users can also provide feedback and ratings to help ChatGPT improve and learn from its mistakes. OpenAI hopes that ChatGPT can become a useful and engaging tool for users, and a valuable source of data and insights for researchers and developers.
