Skip to content
Fast Company Executive Board

The Fast Company Executive Board is a private, fee-based network of influential leaders, experts, executives, and entrepreneurs who share their insights with our audience.

Exploring GPT-3’s capabilities in research

Few inventions in AI have managed to take the world by storm quite like GPT-3 has.

[Image: Adobe Stock/metamorworks]

Profile photo of Saikiran Chandha

Few inventions in AI have managed to take the world by storm quite like GPT-3 has. Upon its public release in 2020, it pretty much immediately captured the attention of every tech enthusiast, computer scientist, and investor. After all, when you’re backed by the likes of Microsoft and Elon Musk, spend millions of dollars, and feed a whopping 45 TB worth of text data to train one of the most powerful large language models (LLMs) ever created, you’re bound to generate some buzz.

OpenAI has been very clear about its intentions: benefiting humanity through artificial general intelligence. The biggest testimony of their motto is that GPT-3’s API is open access, meaning anyone can use its codex to develop a new AI tool. 

So, when the company shared its brainchild, GPT-3, with a world that revolves around innovation, it shouldn’t be surprising to see an increase in proficient AI tools over the past couple of years. 

GPT-3 works on a simple principle. You feed it prompts; it generates human-esque text. The more detailed the prompt, the richer the output will be. There have been several applications of the LLM, such as generating text, music tabs, art, and code, among plenty of other novel content.

But most of these tools cater to the masses. The focus of this article, however, is looking at ways in which GPT-3 is benefitting those who are helping humanity progress: researchers.

WRITING THE FIRST DRAFT

A while ago, a researcher asked GPT-3 to write an academic paper—about itself. And it did a pretty decent job in just under two hours. The paper included correct formatting and necessary elements like references. Would it pass the peer review process? Who knows. But for an AI to write a full paper is impressive.

GPT-3 isn’t a perfect writer, though. It recognizes words but doesn’t completely understand the meaning behind them. This can be seen in some of its trivial or unconvincing outputs.

So while researchers know that it’s not going to give them a polished paper, ready to be published, and journals could eventually crack down on AI-generated manuscripts, GPT-3 has still proven to be more than competent at generating first drafts and breaking through writing blocks.

SUMMARIZING PAPERS

Academicians are bound by time constraints, especially when they’re balancing teaching, studies, and research. And literature reviews are the most time-consuming endeavor when conducting research. Thus, AI-generated summaries from tools like Semantic Scholar’s TLDR are helping researchers save a lot of time. 

Make no mistake: Summaries are no substitute for reading a paper in its entirety. But these quick, brief summaries let researchers sift through irrelevant papers and invest more time in those that suit their research.

PROVIDING CONTEXTUAL LEARNING

Studies show that research papers are getting less readable by the day. It’s not a reflection of the authors’ writing skills but of how technical many papers are becoming since there’s little that can be done about jargon, math, and acronyms.

Traditionally, someone who comes across such a paper would spend time looking up the author, mail them queries, and hope the author wrote back answers. Our company, SciSpace, created a GPT-3-based AI tool that is attempting to solve this problem by giving quick explanations of text, math, and tables in a research paper. Moreover, users can also engage in Q&A to get further context or change the output language in case they’re non-native English speakers. 

ASSISTING WITH CODING

GPT-3 recognizes and predicts almost all forms of text, including code in CSS, JSX, Python, and more. Based on the same principles as text generation, GPT-3 completes codes or generates fresh ones as outputs to written prompts.

Quantitative researchers often work with massive datasets, and analyzing them manually is no piece of cake. While plenty of researchers and Ph.D. students learn to code, AI tools like the one released by GitHub speed up the process of data collection and analysis. 

A FEW MORE THINGS

Of course, GPT-3’s use cases aren’t limited to the applications above. It’s a diverse tool that is being implemented in new ways every day. People use it to convert plain English into LaTeX math equations, create charts and bar graphs, and translate entire papers into different languages.

A BRIGHT FUTURE

By no means is GPT-3 the end-all-be-all of artificial general intelligence. There’s still a long way to go in solving AI. But to think that the first AI model was limited to completing a game of checkers, it’s amazing how far we’ve come.

OpenAI is already in talks of releasing GPT-4, which is expected to be an even bigger LLM than its predecessor and be trained on a much wider dataset, leading to more reliable outputs. 

So far, we’re already creating text, code, and images with GPT-3. Who knows what more is brewing?


Saikiran Chandha, CEO & Founder of SciSpace—the only integrated research platform to discover, write, publish & disseminate research papers.