Web 3 HR: How can DALL-E and GPT-3 help HR?
Two fantastic algorithms are revolutionizing content production around the world: GPT-3 and DALL-E. Haven’t heard of these two gems yet? It’s time to catch up, and understand the implications ahead for the world of content production, as well as for Human Resources.
What is the GPT-3 algorithm used for?
First of all, it must be understood that GPT-3 is neither a platform nor a machine, but a machine learning algorithm based on a dataset of more than 3 million US patents. It was created in 2017 to generate data-driven predictions, and has been continuously refined and improved ever since. With this algorithm, it is possible to generate new content from the knowledge available on the Internet. This is an oversimplified explanation, but it gives a good insight into the power of GPT-3, and the vision of its creators: Google Brain and Open AI.
Google Brain is a deep learning research team at Google. Their goal is to create intelligence that can power everything from Google search and Street View to the latest advances in robotics and self-driving cars. OpenAI is a research company that focuses on artificial intelligence (AI) to promote user-friendly AI. The company was founded by several co-founders, including Jack Hughes (co-founder of Akamai Technologies) and Elon Musk (founder of Tesla, SpaceX and several other startups). OpenAI’s goal is to “advance digital intelligence in ways that are most likely to benefit humanity as a whole.”
The GPT-3 algorithm is thus already widely used by large technology companies such as Google, Facebook, Microsoft and IBM. The main use cases of GPT-3 are:
- Generate text, either for general purposes or for specific tasks such as machine translation or summarization.
- Generate training data for machine learning models.
- Serve as a knowledge base for artificial intelligence applications.
The settings used by GPT-3 vary depending on the specific application or task for which it is used. However, some of the most common parameters used by GPT-3 include the size of the training dataset, the number of hidden layers, the number of neurons in each hidden layer, the learning rate, and the activation function.
To allow you to understand more concretely what this algorithm is for and how it can be used, I suggest that you read this chapter again. Indeed, it was entirely written with the assistance of GPT-3. I asked him a few simple questions, for which he generated original texts for me. The algorithm is able to generate full paragraphs within one or two seconds, and then it took me about 5 minutes, to modify sentences here and there. Something to think about the future of content production!
The principle is simple: you enter a description of what you want to detail, like ‘why can’t we accept death?’. Note that the answer is creepy because it is self-generated: ‘Death is a difficult thing to accept because it is the end of life. It is the end of a person, and it can be hard to let go.’ This text was completely generated by the algorithm. That is to say that this text is unique, and has not been copied to another site to bring it out to you, as may be the case for a Google search.
Now that the sandbox version is available to the general public, you can go try it out for yourself:
As you will see, there are many parameters you can play with to fine-tune the quality of answers: desired length, originality, relevance, accuracy, etc. The basic algorithm is the Da Vinci version, but if you are an expert, you can use the API to play with more specific models for certain tasks, such as comparing and classifying texts or documents, reading pdf or images, etc.
What is the DALL-E algorithm used for?
The operation of DALL-E is very similar to that of GPT-3, except that it allows to generate images. It is really very surprising. From a description of what you want to see, the algorithm generates an image. You will surely not have missed this incredible octopus above. Believe it or not, it was created by DALL-E from the description I gave him, which was: ‘a photo of a purple evil mechanical octopus with red eyes floating in purple electric water’.
Having access to the beta version, I have a lot of fun generating images. With this technology, I think Shutterstock and other Pixabay will quickly get out of my hangouts… For example, what description do you think I entered to get the two variations of the image below?
This is what I typed: ‘An expressive photo of a basketball player dunking, depicted as an explosion of a nebula’. You will notice the expressive style, very artistic. Above all, the algorithm generated an image of a man and an image of a woman, since I didn’t specify gender. I don’t know what you think about it, but it looks like a new art form!
A few weeks ago, I was at a L’ADN media evening, where a digital artist presented some of his works made with an algorithm similar to DALL-E. When you don’t know how to draw, paint or photograph, you can now describe what you would like to create, and the algorithm can convert this description into an image. With DALL-E, it is possible to specify even the desired style. So if I change the art style from ‘photo’ to ‘pixel art’, this is what I get:
You will notice that beyond the style, the images are not exactly the same. This is simply because DALL-E creates the entire image on every request. If you generate the image of the same description ten times, you will get images that are close, but never similar! I intend to enhance my articles, newsletters and picture books created on DALL-E from now on…
I wish I could give you access to the sandbox, but it’s still invite-only at the moment. As soon as the public beta opens, I’ll make you bite. In the meantime, you can learn more on the public site, and feast on images thought up by platform users!
GPT-3/DALL-E & RH: what use cases?
GPT-3 and DALL-E can be used for human resources in several ways. For example, GPT-3 can be used to help identify potential candidates for open positions, to help assess candidate qualifications, and to help match candidates with open positions. Additionally, GPT-3 can be used to help create and administer training programs for employees, and to help evaluate employee performance. It can also be used to create relevant recruiting and employer branding content.
Creation of job offers
GPT-3 can be used to write job descriptions by first understanding the job requirements and then using its natural language processing capabilities to generate a description that accurately reflects those requirements. Here is a real-world example generated for the Data Analyst job:
What are the twenty main missions and tasks of a data analyst?1. Collect and analyze data to identify trends and patterns.
2. Develop data-driven solutions to business problems.
3. Design and implement data-driven decision models.
4. Conduct statistical analyzes to support decision making.
5. Develop data mining models to uncover hidden patterns and relationships.
6. Prepare reports and presentations to share findings with stakeholders.
7. Collaborate with cross-functional teams to implement data-driven solutions.
8. Manage and maintain databases and data warehouses.
9. Monitor data quality and integrity.
10. Develop and implement data security and privacy policies.
11. Troubleshoot data issues.
12. Stay up to date on new data analysis methods and tools.
13. Participate in data-related projects from start to finish.
14. Understand the business context of data analytics projects.
15. Identify opportunities for improving data-driven decision-making.
16. Communicate findings to stakeholders in a clear and concise manner.
17. Write code to automate data analysis tasks.
18. Create dashboards and visualizations to communicate data insights.
19. Use data to inform business strategy and decision making.
20. Help organizations make data-driven decisions at all levels.
Having tried it on more than thirty different professions, I must admit that the result is stunning. The description is all the more precise when the brief is clear. For example, instead of just asking for a job description, I ask for the top twenty assignments and tasks.
Creation of employer brand content
If GPT-3 can effectively describe trades, it can do the same with companies, especially if they have an online presence. Below, I asked the algorithm to list me the 20 specific strengths of LVMH employer brand:
What are the 20 specific and unique strengths of LVMH employer brand?1. The history and prestige of the LVMH brand
2. The company's emphasis on luxury and quality
3. The solid financial performance of the company
4. The global reach of the company
5. The company's commitment to sustainable development
6. Company focus on customer experience
7. The company's focus on innovation
8. The company's high employee retention rate
9. The diversity of the company's employees
10. The company's commitment to employee development and training
11. Competitive company salaries and benefits
12. The company's generous employee discount
13. Company employee recognition and reward programs
14. The company's solidarity and collaborative culture
15. The company's commitment to social responsibility
16. Company commitment to diversity and inclusion
17. The company's global network of luxury boutiques
18. The company's partnerships with top fashion designers
19. Corporate sponsorship of major fashion events
20. The company's flagship store on the Champs-Élysées in Paris
If this work were to be done by employees, it would probably take a few hours of brainstorming type workshop. GPT-3 responded in less than 3 seconds. Of course, it’s certainly not perfect, but we can say that the arguments put forward by the algorithm are convincing to say the least.
Other use cases requiring training of the algorithm:
If the first two examples can be made with the generic version of GPT-3, it will be necessary to parameterize much more finely and to train the algorithm for all the following use cases:
- Screening and evaluation of candidates: from the CV of the candidates (CV parsing), the Internet and detailed information on the position to be filled, GPT-3 is able to determine a priori compatibility on the position, and to identify and rank the best candidates.
- Creativity assessment: Thanks to DALL-E, it is entirely possible to develop a psychometric assessment that measures a person’s creativity along several dimensions. We could thus determine the originality and relevance of a description, and its link with the created and refined work (DALL-E makes it possible to generate successive variations of an image).
- Assessment of written expression and synthesis skills: thanks to GPT-3, it is quite possible to develop a psychometric assessment to measure a person’s ability to synthesize and express. To get detailed and relevant content, you have to know how to use the right words and be able to articulate them perfectly. From pre-tested use cases, it would be possible to identify performance levels.
- Creation of training programs and content: GPT-3 having assimilated a great deal of knowledge, it can play both the role of Google Search, and that of an educational engineer capable of presenting the knowledge in a specific context. This makes it possible to create entire libraries of relevant and engaging content.
- Employee Performance Review: This last use case is more complicated, but definitely possible. By training GPT-3 with enough internal data on employees and the life of the company, it becomes possible to determine performance levels, and to automate the processing of data related to the achievement of objectives.
Some of these use cases may work with little to no customization of the algorithm. On the other hand, for the others, it will be a question of choosing the most appropriate version, then of training it with a dataset specific to the subject to be treated. Regarding the versions, there are now about ten, grouped into families: Da-Vinci, Curie, Babbage, Ada. Beyond GPT-3, there are independent and specialized algorithms, in particular used by AI artists, which can freely define the parameters for interpreting the data.
It is obvious that to appropriate these technologies, it will be necessary to have a strong expertise in data technologies. This will involve having a team available internally, or calling on an expert web 3 HR firm like Tomorrow Theory.
GPT-3 & DALL-E: ethical and humanistic considerations
These innovations are in no way essential or necessary in our current daily lives. They are simply part of the logical evolutions of our society, which constantly seeks to optimize the efficiency of tasks. If there is a solution so that content can be created cheaply in seconds, against several hours of a paid human, it is certain that our society will eventually take this path. No value judgment, it’s a pragmatic statement. We can rejoice or be sorry, but that’s another matter.
Some will see in these advances a reduction of the human by the machine. On the other hand, some will find themselves able to express themselves with a quality of graphic or textual expression that was previously not accessible to them. There are obviously education and safeguards to put in place, but the value brought by DALL-E and GPT-3 is undeniable. Personally, not knowing how to draw or paint, I take great pleasure in refining my expressions to create visuals that represent my thoughts.
It is interesting to note that the algorithm has been restricted so as not to allow the creation of content that is a priori immoral. War, nudity and other politically incorrect concepts are blocked. This obviously raises the question of who creates these rules and these limitations, and I dare to hope that in the near future there will be a governance that is clear and as inclusive as possible.
In terms of HR processes, it will be easy to fall into excess, as for the example of the automated evaluation of employee performance. GPT-3 was designed to compare, classify and interpret, so make no mistake, it would actually be quite simple to unlock this use case. It will then take enlightened and virtuous leaders and HR not to do anything with this new power…
On the economic side, GPT-3 and DALL-E will force text and image producers to become more original and efficient, so as not to be replaced. Personally, I think it’s logical and relatively virtuous, even if I understand the opposition. There will always be room for original creatives, capable of offering qualitative and expert approaches. For example, on my HR Tech subjects, I am not worried about seeing my content fully replaced by AI, as it is a fairly uncommon expertise, which I approach with a multi-disciplinary approach including technology, psychology , philosophy, history and economics.
GPT-3 & DALL-E: A New Content Paradigm
Few people in France are still well informed about GPT-3 and DALL-E, and even fewer people have had the opportunity to play with it. Even if you will have to wait a little longer to test DALL-E, you can already create incredible texts in English with GPT-3, which you will then only have to translate into French with Google Translate.
And if you think these AI algorithms have nothing to do with web 3, then you’ll have to dig a little deeper into what web 3 is. As a reminder, web 3 can be summed up as technological convergence blockchain, virtual reality, and the spatial web. This last block notably contains artificial intelligence technologies such as semantic processing. So GPT-3 and DALL-E are on the contrary essential bricks of web 3, which will allow users to generate original and relevant content effortlessly. A whole new world…
Enough to say that the students will have a field day writing their essays. Also, it is clear that GPT-3 paves the way for a new quality standard in the processes of the HR function. What we can remember for the moment is that with a little creativity and common sense, these technologies have the ability to profoundly change the HR tasks related to the creation of content.
This is an interesting gateway to web 3, and at Tomorrow Theory, we look forward to supporting the HR function in getting to grips with these tools for real virtuous innovations. And for my part, I will try the experience of writing and illustrating an entire book next year thanks to GPT-3 and DALL-E… To be continued!
— —
If you are interested in the combination of web 3 with HR, I invite you to subscribe to the dedicated newsletter that I keep writing on the subject, and to read the articles that I have written on the topic:
