Magazine : 
ChatGPT- 4: A rise or fall of Artificial General Intelligence
OpenAI has quickly become one of the biggest names in tech. The artificial intelligence (AI) company has made realistic image generators, 3D-model creators, and the thing it is now best known for, ChatGPT. In less time than it takes me to write this sentence, ChatGPT, the free artificial intelligence computer program that writes human-sounding answers to just about anything you ask, like explaining quantum physics with literary flair. But recently, the latest and most advanced version of its language model, GPT-4, was introduced, which also showed signs of artificial general intelligence (AGI). According to a preprint paper titled Sparks of Artificial General Intelligence: Early experiments with GPT-4, the model demonstrated abilities that were at or above the human level in a range of tasks, including mathematics, coding, vision, medicine, law, and psychology. The paper also showcased GPT-4's ability to write proof about the infinite number of primes, complete with rhymes on every line, and draw a unicorn in a drawing program.
If you ask what the difference between ChatGPT and GPT-4, is ChatGPT is more like a car and GPT-4 is the engine behind it, a general technology that can be used in various applications, including language learning, chatroom monitoring, and assistive technology. Compared to its predecessor, GPT-3, GPT-4 performs better in technical challenges, such as answering math questions and avoiding false answers. It also has a sense of ethics built into the system to prevent it from performing malicious or harmful tasks. OpenAI has released a long paper of examples of harms that GPT-4 has defences against, including the ability to decline tasks like ranking races by attractiveness or providing guidelines for synthesizing sarin, a toxic gas.
One of the most significant changes to GPT-4 is its ability to handle over 25,000 words of text, making it ideal for long-form content creation, document search and analysis, and extended conversations. Additionally, GPT-4 can pass a simulated bar exam with a score around the top 10 percent of test takers, making it an invaluable tool for professionals in various fields. And other aspects of GPT-4 are its ability to understand and analyze images. GPT-4 is more advanced than Google Lens, as it can analyze an image and provide a detailed explanation. GPT-4 also outperforms previous versions when it comes to multilingual capabilities. OpenAI has demonstrated that it outperforms GPT-3.5 and other LLMs by accurately answering thousands of multiple-choice questions across 26 languages. While it handles English best, Indian languages like Telugu are not far behind, making it an invaluable tool for users in non-English-speaking countries.
It Comes with Limitations
However, the Microsoft researchers who authored the paper quickly dialled back their initial claim, acknowledging that GPT-4 still had significant limitations and biases. They noted that their definition of AGI was based on a 1994 definition of intelligence by a group of psychologists, which defines intelligence as a very general mental capability encompassing a broad range of cognitive skills and abilities. While GPT-4 showed progress in these areas, it still needed to catch up to many of the traditional definitions of AGI.
OpenAI CEO Sam Altman also emphasized the limitations of GPT-4, stating that it was still flawed and limited and required a lot more human feedback to be more reliable. He noted that while OpenAI was focused on building AGI in the future, GPT-4 was not AGI. Altman warned against the hype surrounding GPT-4, stating that people were begging to be disappointed and that the model was not a substitute for human effort.
The limitations of GPT-4 highlighted by the Microsoft researchers included challenges with confidence calibration, long-term memory, personalization, planning and conceptual leaps, transparency, interpretability and consistency, cognitive fallacies, and irrationality. While GPT-4 showed impressive capabilities in various tasks, it still needed help with basic cognitive abilities that humans take for granted, such as knowing when it is confident or just guessing.,The debate surrounding GPT-4's capabilities underscores the ongoing challenge of defining and developing AGI. While GPT-4 represents a significant advance in language modelling and natural language processing, it still falls short on many of the traditional definitions of AGI, which require a machine to exhibit not only a broad range of cognitive abilities but also consciousness, intentionality and other human-like traits.
Despite the safeguards and built-in sense of ethics, some worry that teaching an AI system the rules may also lead it how to break them. Dubbed the Waluigi effect, this outcome occurs when tricking the system into deciding not to be ethical, and it will merrily do anything asked of it. While GPT-4 has been designed to prevent such outcomes, researchers have demonstrated that it is still possible to simulate malicious behaviour with the system.
In conclusion, GPT-4 is an impressive language model that has progressed in various tasks but still needs to be AGI. While it has built-in safeguards to prevent malicious or harmful schemes, concerns remain about potential unintended consequences. The hype surrounding its capabilities should be tempered by clearly understanding its limitations and biases. As AI continues to advance, it is essential to remember that developing AGI is a long-term goal that requires not only technological breakthroughs but also careful consideration of such a powerful technology's ethical, social, and political implications.
Read More News :
Trustt Launches First Major GPT Product Suite for Banking Industry
