ChatGPT & the future of AI

The launch of ChatGPT’s beta version exceeded a million users in less than a week and the buzz is still out there. The launch of this platform by OpenAI attracted the attention of the entire tech ecosystem with featured articles in major publications. All of this with just one tweet by Sam Altman.

ChatGPT’s buzz is not only about the leap in the field of artificial intelligence but it also garnered the buzz of the possibility that it might pose a threat to human jobs. Jobs such as creative writing, legal document drafting, and customer services inquiries to name a few. But is it the reality?

Let us answer the question. There are more nuances to how we see the potential applications of generative AI like ChatGPT and in a broader spectrum, Large Language Models, LLMs, where the reliability of information is of prime importance. We discussed with industry experts and our team, how ChatGPT can be used in conjunction with ML algorithms for workplace automation.

When considering applying ChatGPT or any other generative AI model such as Dall-E, veed.io, etc, it is important to consider the fundamental limitation. The limitation is that generative AI models do not read sources or cite work, they simply generate responses & therefore, the output has no guarantee of reliability.

To use generative AI models within organizations, it is necessary to feed these models with cited information & facts so that the answers are not inferred but are simply based on research.

To further understand this, let us look at how ChatGPT was trained.

A large database of information which includes user-generated data, such as social media posts, chats, memes, and similar other things was fed into the ML algorithm. Now, if a company wants to use ChatGPT for a specific purpose, data that is specific to its area of interest would need to be obtained to train the model further. For example, a bank would need to use chats between their customers and relationship managers.

This type of data is usually difficult to obtain & some companies choose to generate such data. DesiCrew has expertise in generating training data for clients based on specific requirements. We helped one of our clients build an automated translation tool for Indian languages. Automated translation is one of the tricky AI tasks given the fluidity of human language. The product is currently being used by 200 million users.

If we look at the data annotation aspect, many methods will be required to prepare the training datasets. As an example, annotators will have to perform text classification, a process of assigning tags to different parts of the text to organize, structure & categorize the data. Another thing necessary is entity annotation which will involve locating, extracting & tagging entities in the text.

DesiCrew is a global company for data annotation outsourcing, trusted by many GAFAM & Fortune 500 companies as well as innovative startups. With 16 years of experience, our team stands strong with 1500+ professionals helping businesses with their advanced data annotation challenges. Reach us at [email protected] to know more.