Not hearing of the name ChatGPT may come as a shocker, considering the enormous popularity of the platform in the past few months. Launched in November 2022, the platform has been garnering news around the world and has been making some great headlines.
If we take a recent example of how the AI tool has answers to everything in the most humane way, we cannot overlook this emotional answer to a query by a Twitter user named Cynthia Savard. She asked the chatbox to explain to her children in a letter that Santa is not real, and here is what the chatbox had to say:
What is ChatGPT?
It is a new-generation chatbox that translates all your queries into an appropriate form of information in real time. The chatbox is built over the language model GPT-3 (Generative Pretrained Transformer) which uses deep learning. In turn, it provides human-like text.
Also Read: 5 Best Apps To Buy Cryptocurrency In India!
Since the day ChatGPT’s prototype was launched, people have hailed it for unimaginable human-type answers that are generally unavailable with other frequently used chatbots. The models have been perfected in collaboration with Microsoft using their Azure supercomputing.
Why is ChatGPT famous?
As discussed in the previous sections, the growing popularity of ChatGPT is based on the fact that it is able to produce human-type answers really quickly. In fact, there have been observations that one can pass a law course, medical, and other examinations using ChatGPT with slight human intervention. This comes as a huge relief to different stakeholders who earlier did not have access to AI-powered chatboxes like it.
Additionally, it caters to a wide range of knowledge based on multiple domains. Imagine getting essays, assignments, codes, and even getting answers to your examination question written for you so that you do not have to do any additional tasks. Yes, ChatGPT does that for you!
The ChatGPT platform is based on text databases that are collected all over the Internet based on searches, and other documents present. In this section, we take a brief idea of how it works and the basic flowchart of the working.
- Built over GPT-3 model and aligned with human feedback to give exactly what the person wants. In the earlier model of GPT-3, the answers did not meet the human expectation and there was an inherent lack of clarity and interpretability.
- Reinforcement learning based on feedback from humans is used in the model.
- The initial process is Supervised Fine Tuning (SFT) which gathered prompts suggested by humans as well as those submitted to OpenAI on their platform. The answers were gathered by human beings and created a small yet fine set of data.
- Next in the series of steps was the use of the Reward Model. One prompt from the SFT model generated multiple answers to it. Human labellers now ranked the data presented from best to worst to give it a human touch and preference.
- The last among the list of steps is using Proximity Policy Optimisation (PPO) along with the SFT model to better the outputs. Based on the prompt and output, it gives a reward-based answer to the user.
Based on the parameters of greater helpfulness and lesser toxicity, the answers were evaluated and then displayed.
ChatGPT as a revolution
Even though there were AI-powered chatboxes beforehand too, there was never something as powerful as ChatGPT. The plain and simple answer to this is due to the involvement of human feedback. Answers are accurately presented based on what humans predicted.
For earlier language-based chatboxes like GPT-3, the problem was an inherent lack of clarity and truthfulness. In one of the recent surveys conducted by Microsoft, 85 per cent of respondents voted for automation technology in workplace, on the line of ChatGPT.
Even though the platform comes laden with multiple positives, one cannot contradict the negatives that the world may see propping. Here are some of the implications of the AI-powered chatbox and how it impacts multiple stakeholders.
- The rise of tools like it has been construed by many educationists, teachers, and others as dis-incentivizes originality.
- Students may retort to plagiarism and unethical data content that is copied and pasted from the platform.
- One does not have to spend long hours studying textbooks and resources to complete projects and assignments. ChatGPT does that for them and hence tries to bypass the regular studying methodology.
- More often than not, there always has been an issue with producing vast amounts of content for websites. This is to ensure that it ranks higher on Google searches.
- Such AI-based models may reduce the role of authentic content. Instead, there would be more emphasis on quantity over quality.
- Agencies would not require content developers to do the job for the client. ChatGPT does the same for you at a reduced cost and timeline.
News and Publishing
- In a world already grappling with the issues of fake news or inauthentic information, the era of ChatGPT and its impact on the news industry might be regressive.
- Additionally, plagiarised content may further cause problems for journalists and media agencies.
- Inaccurate information and political propaganda could even lead to violence and hatred in society.
Keeping jobs alive(recession+position cuts due to automation)
- The impact of impending recession is already visible in the global economy.
- The job cuts, lay-offs, and freezing of hiring across domains will already reduce employment across the world.
- With automation from tools like such, there may be further resource cuts and lesser allocation to sectors whose work can be easily done with AI tools.
Also Read: Top 5 Stock Trading Apps In India!
Here’s a recent interview of Cashify TechByte with Rishabh Prakash, a Senior Data Scientist, who spoke to us at length about ChatGPT, its application, evolution, and expectation from the language model.
Q- Do you think that ChatGPT is an influential player in Industrial Revolution 4.0?
A- Industrial Revolution (IR) 4.0 is a combination of different parameters like Big Data, the Internet of Things, etc. We even call it Web 4.0 due to the use of artificial intelligence (AI), machine learning, and similar application. ChatGPT is a language model that is trained on a vast amount of data and derives itself from the GPT 3 model. Basically, it is based on the conversation between the user and trainer. Depending on the answer, a reward is provided. If I take the example of food delivery applications then it can use this newly launched model to keep a tab on the previous order of customers. Hence providing personalised recommendations, can bring a great leap. It can be a great player in IR 4.0 when combined with parameters like the internet of things or big data.
Q- Since human labellers are involved in some of the steps, do you think that with such human interventions, unbiased answers are possible with the platform?
A- ChatGPT is based on the dataset that has been fed into it, and it does paraphrasing to provide unique answers. However, I think that with human involvement, there can be problems of getting biased answers. For example, there have been cases where people have felt that it is giving more liberal and woke solutions aligned to some political or regional sentiment. This can be offset by increasing the number of datasets from multiple sources. Ultimately, this will neutralise the overall impact of any biases.
Q- Are there problems of rigging, too, that one can speculate in times to come?
A- I believe that the existing issues such as fake news, morphed images, etc., the problem of rigging can exist if the data sources for the text generation platform are limited. The best way out is to increase the data set from multiple sources. In the case of social networking sites like Facebook, they already have a system in place to ensure that the data placed in the public domain is verified. Even though it takes some time to validate the data manually, it still ensures that only accurate and original data is provided to the user. However, until it is removed (in case of inauthentic or fake news), people have already taken screenshots or recorded the existing video, which keeps circulating over and over again. ChatGPT can improvise more ethical standards and policies to minimise such problems.
Q- The rift between genuine and AI-generated content is a big challenge that has come up since ChatGPT. How do you think that search engines are going to cope with this and maintain the ranking based on quality?
A- Search engines can employ the use of AI as well as manual reviewers. The latter can actually review the content that is placed on the website. For example, when you generate content manually, there are a few inconsistencies. This can be in terms of grammar or the tone of the content. However, with ChatGPT, that inconsistency is removed. The human-generated content can be differentiated by manual reviewers as well as AI. Besides, ranking is always based on user crawling. If there is lesser human/ natural language, then people might not crawl more towards this content. Ultimately, it will affect ranking.
Q- Do you think ChatGPT will replace manual jobs with automation?
In terms of automation eating up jobs etc., similar speculation came when the computer-based service industry was launched. At that time too, people thought that it might end up taking away jobs of people. However, the case was much different. Many people got employment. In my opinion, with the use of the platform, there might be a higher application for it to improve existing services, and people might have more avenues to get jobs too.
Q- What, according to you, is one of the ‘never found before’ features of ChatGPT that will be a big game changer in times to come?
A- According to me, features like text generation, sentiment analysis, language translation, chatbots, content creation, and summarisation are already a game changer. Additionally, it can be used for generating news for the media domain, maintaining consistency, and for educational purposes for assessment.
I think one feature that can be a big game-changer is content creation. When used along with other AI applications, it can yield some amazing results. This can be helpful for industries like marketing and ad agency or education. Try combining text-generating AI with image-generating AI like Dall-E, it can complete the task comprehensively for you.
- New York City (NYC) education department has already banned the tool in all the local networks and devices associated with the schools.
- Stack Overflow (Q&A website for professional and enthusiastic coders and computer lovers) has banned ChatGPT. This is due to the inaccurate and misinformed answers it produced.
- Bangalore-based RV university in India has put an end to students using the platform to complete assignments, practicals, and other examinations.
- French University Sciences Po has banned the platform over a plagiarism row that arose with the students’ submission at their campus.
Considering one of the major involvement of human labellers to model the ChatGPT output, there may be chances of biases and interference in the final answers. However, one cannot expect it to replace everything in existence. For example, even though AI tools can help students make assignments and write projects, the long-term repercussions of the tool can be well taught to them by teachers and schools.
Google itself is in work with an algorithm that looks to restrict and rank the content produced by AI tools accordingly. This will ensure that original content will always be ranked better and higher.
One who turned her passion into a profession. Traveller by heart and minimalist by nature, I am a people person and voice to the voiceless. A Bollywood buff and tech enthusiast who intends to cover the nooks of the world.