CHAT GPT 4
Introduction:
- OpenAI, a US company, just announced GPT-4, its newest AI model. A more advanced version of the company’s chatbot, ChatGPT, will be powered by this vast language model, which has the ability to understand and generate inventive and clever discourse.
The capabilities of GPT-4:
- GPT-4 is a significant improvement over GPT-3.5, which introduced ChatGPT.
- More imaginative and communicative than GPT-3.
- Its most important feature is its capacity to simultaneously gather text and visual input and take into account both when generating a response. According to reports, the model can recognise human emotions in photographs, including humour. Its ability to describe visuals already benefits those who are blind or visually impaired.
- GPT-4 was put through a battery of human-designed tests, and it performed noticeably better than usual. When compared to its predecessor, which only scored a 10% score on a mock bar exam, it performed in the 90th percentile. GPT-4 also excelled in advanced classes in environmental science, statistics, art history, biology, and economics.
- On the other hand, the GPT-4 only received a 40% on the advanced English language and literature sections. In terms of language understanding, it exceeds other high-performing language models in English and 25 other languages, including many Indian languages.
- Since its release, ChatGPT-generated content immediately made an appearance in high school and college essays; at the moment, this tool’s capacity relates to testing platforms.
- GPT-4 can perform a variety of white-collar tasks, especially those involving writing and programming, according to preliminary numbers from OpenAI, but it generally ignores tasks involving manufacturing or science. The economy and public policy will be impacted by a broader deployment of language models.
- Planning and learning still need to be mastered, but GPT-4 already satisfies four of the seven criteria for intelligence, which is defined as “a highly general mental skill that, among other things, encompasses the ability to reason, plan, solve problems, think abstractly, absorb complicated ideas, learn quickly, and learn from experience.”
Ethical conundrums:
- There are still a few issues with GPT-4’s predecessor. Its output might not always be factually accurate, a quality that OpenAI refers to as “hallucination.”
- There are some unfavourable preconceptions and biases in the Internet-scraped content used to train GPT-4. Another presumption is that a large dataset will be diverse and fairly representative of the entire planet. Yet, individuals from economically developed countries, young individuals, and males with manly voices predominate online.
- The moderator model only picks up on the biases we are aware of and those that primarily influence English. Preconceptions that are prevalent in non-Western countries, such as those based on caste, could not be known to this model.
- The misuse of GPT-4 as a tool for spreading false information and propaganda is very high.
- The most crucial question here is whether the decision to desist from acting incorrectly should come from the human mind or the computer’s guidelines.
“Unpredictable parrot”:
- In its most basic version, GPT-4 is a computer software that uses probability it has learned through training on huge text corpora to predict the next word in an unfinished sentence. GPT-4 is often characterised to as a “stochastic parrot” since it can speak in understandable terms without understanding their meaning, however according to Microsoft Research, GPT-4 understands what it is saying and not all intelligence is a type of next-word prediction.
- In addition to OpenAI’s models, ChatGPT rival Claude has been presented by AI company Anthropic. Google has unveiled PaLM, a model that is instructed to have more degrees of freedom than GPT-3.
The UPSC CSE is not intended for machines:
- Recent reports claim that ChatGPT “failed” the UPSC Civil Service Examination (CSE) Prelims. By asking it to respond to questions from the UPSC Civil Service Prelims 2022 exam, the AI chatbot was put to the test. Just 54 out of 100 questions could be accurately answered by ChatGPT, though.
- It illustrates the challenges of the UPSC CSE, one of the most challenging exams in the world, as well as the limitations of machine learning.
Conclusion:
- Global efforts are being done to develop a model with a trillion degrees of freedom in general. These will be incredibly large language models that raise worries about what they cannot do, but these worries would be red herrings that would divert our attention away from the question of whether we should be developing models that merely test the limits of what is feasible without taking society’s concerns into account.
