GPT-3 and what it means for recruitment

There has been an explosion in conversations around GPT-3 technology of late. This is due to the recent release of ChatGPT, which took place last month.

If you haven’t come across ChatGPT yet, it’s getting a lot of attention from passive as well as passionate tech observers due to its ability to answer questions, write articles, and code. But is there an application of this technology in recruitment?

Understanding the technology

Well, before we go any further, we need to understand what this technology is and how it works. It’s worth mentioning that, while both were developed by OpenAI – a San Francisco-based artificial intelligence research laboratory – GPT-3 and ChatGPT are not the same. However, they are based on the same Generative Pre-Trained Transformer architecture (GPT).

According to ghacks.net, both ChatGPT and GPT-3 use large amounts of data and powerful computational resources to learn the patterns and structure of language.?

ChatGPT, though, has been optimised for more chat-like interactions. As a result, it’s far more user-friendly – which is why its release has garnered far more mainstream enthusiasm than the release of GPT-3 back in June 2020.?

So that’s GPT-3 explained in the context of its differences from ChatGPT. But for an explanation of exactly what the source technology is, it seems fitting to use the technology itself to explain. When asked ‘What is GPT-3?’, ChatGPT responded with:

"GPT-3 (short for "Generative Pre-training Transformer 3") is a language generation model developed by OpenAI. It is one of the most advanced language models to date, with a capacity of 175 billion parameters. GPT-3 can perform a wide range of language tasks, including translation, summarization, question answering, and text generation."

Seems pretty clear and concise, but let’s expand on that a little… The ‘language generation model’ is quite simple in itself, however, it is autoregressive, meaning it predicts future values based on past values. In simple terms, it learns as it goes to produce human-like responses.

This being the third iteration in the GPT-n series, it’s more advanced than the previous two. And that’s most clearly demonstrated by its whopping 175 billion parameters*.?

*For reference, parameters are the configurable variables within the machine learning model which enable the estimation of the output from any given input. And for context, the second most powerful language model is Microsoft’s Turing NLG, which has around 17 billion parameters.

Whilst 175 billion may seem like a large number, we’re still talking about Narrow or Weak AI. This is the categorisation for pretty much all current practical uses for AI, including chatbots and recommended viewing on streaming services.?

General uses for GPT-3

Chat

You don’t need to spend too long experimenting with this technology to realise its potential for sales, marketing, customer service and human resources activities. Conversations with GPT-3 are fluid and, once the bot is trained in the specific information and terminology of each use case or industry, it can supplement the work of a human, saving time and money.

Text-to-code problem-solving

There is a major problem-solving application for GPT-3. For basic web design and functionality edits and maintenance, you don’t need extensive knowledge of languages like JSX or PHP. You can just type what you want through GPT-3 in plain text and it’ll give you the relevant answer in code.?

A great example of this text-to-code capability is provided by @sharifshameem, who wrote a plain text sentence describing what Google's home page should look like, and GPT-3 turned into code with incredible accuracy.

This problem-solving functionality can also be put to work on more everyday solutions, too. For example, fixing lines of code within spreadsheets to achieve the desired result.?

How can GPT-3 benefit recruitment?

Now on to the question or questions we’re all really interested in: Can this technology benefit recruitment? And if so, how?

Like with many forms of automation in recruitment, there are myriad time-and-effort-saving benefits. And here are four examples to consider:

Emails

Emails take up a considerable amount of your workweek. And what’s often the most time-consuming aspect of this task is the incremental edits to emails that have the same function, but a different intended recipient. This is necessary seeing as the average worker receives 121 emails per day, and there is a requirement to stand out from the crowd.??

Often, though, despite your best efforts, these emails end up feeling templated – there simply isn’t the time to spend on making every email feel bespoke.

However, if you train GPT-3 using samples of your previous emails it will generate new emails using words and phrases that capture your tone and sound authentic without completely copying the samples you put in.

The automation of emails can also be more broadly applied to general communication with your candidates. For example, responding to queries related to the recruitment process, application status, and interview schedule. GPT-3 can provide you with time-and-effort-saving benefits while still delivering communication that feels personalised and natural on the candidate’s end.

Resume screening

Sorting through applications and CVs is time-consuming. According to Ideal, the average job opening receives 250 resumes. This means you can spend up to 23 hours screening resumes for a single hire.

Whilst there are existing solutions to the screening pain point, they can be bypassed by the scourge of keyword stuffing. This contributes to, as shown in the Ideal data, the problem of up to 88% of submitted CVs being from unqualified candidates.

With GPT-3 you can not only train it on a dataset of resumes and job descriptions to automatically identify candidates who possess the required skills for a particular role, but you can introduce semantic search. This will look beyond the keywords or phrases you enter into GPT-3 and highlight candidates whose resumes include synonyms and tangential skills, too.

Through semantic search, GPT-3-based resume screening will enable you to identify the best people for each role – those that might’ve been overlooked previously. And it’ll also help you resurface hidden candidates in your database.

Diversifying the talent pool

We know that the more diverse the team the more they achieve. A CIPD study shows that diverse teams see a 60% increase in their ability to make decisions. And a report by McKinsey found that companies in the top quartile for ethnic and racial diversity in management were 35% more likely to have financial returns above their industry mean.

These benefits have been recognised by recruiters, too. According to LinkedIn’s Top 100 Hiring Statistics for 2022, 30% of recruiters have specific goals and policies in place around hiring for racial and gender diversity. Despite this, however, 38% of recruiters say finding diverse candidates to interview is the biggest barrier to improving diversity.

That’s where GPT-3 comes in. It could be used to help diversify your talent pool and therefore submit more diverse candidates. You could achieve this by training it on a dataset of resumes and job descriptions to automatically identify candidates who possess the required skills for a particular role, while also taking into account factors such as race, gender, and socioeconomic background.

You could also use it to generate job descriptions and ads that are more inclusive and appealing to candidates from underrepresented groups. Both would ensure engagement with a more diverse range of candidates.

Writing job descriptions

According to HRSG’s State of Job Descriptions 2020 Survey, 65.1% of HR professionals spend over two hours writing job descriptions. It’s a common pain point in recruitment.?

There are many contributing factors to the length of time it takes to create a good job description. Among them are whether the HR professional has been trained in writing job descriptions, and whether they’ve got access to a quality tool to help them.

By training GPT-3 on a dataset of existing job descriptions, it can generate new ones based on that training. The generated job descriptions you get can be highly coherent, grammatically correct, and semantically similar to the training data.?

However, it is important to note, as is largely the case with GPT-3, the quality and accuracy of the generated job descriptions will depend on the quality and diversity of the training data.

Conclusion

Due to their many benefits, automation and AI have become an ever-increasing presence in the recruiting landscape. According to LinkedIn’s Top 100 Hiring Statistics for 2022, 68% of recruiting professionals say that investing in new recruiting technology is the best way to improve recruiting performance. And it certainly looks as though GPT-3 falls under this umbrella.?

With the release of GPT-4 on the horizon, this year could see the use of this technology in recruitment increase exponentially.