Large Language Models: Transforming the Future of Work

Amsterdam-based Monique van Dusseldorp is a busy events professional, curating conferences on technology, media, and innovation. Researching new event and panel ideas has always been a part of her working life.

But recently, van Dusseldorp has been making use of a new way to conduct that research. On a busy morning in November, she sat down at her keyboard and typed a question: ‘Can you come up with an immersive experience that brings people together, based around the world of the famous illustrator Roland Topor?’

Soon, text started to flicker over the screen. The answer came back:

I propose a 'Surreal Garden Walk' in a public park or botanical garden. The garden would be transformed using sculptures and installations inspired by Topor's artwork, creating a whimsical and slightly eerie atmosphere. Hidden speakers will play surreal soundscapes, and sculptures will subtly move or change as people approach them.

Van Dusseldorp continued: ‘What kind of media and technologies can help create an immersive experience?’ The reply:

Think about Spatial Audio: Utilizing 3D sound technology to create a realistic and immersive auditory environment, where the audio changes dynamically with the user's movements.

This isn’t text on a web page, or an answer from a colleague in an industry forum. Rather, van Dusseldorp was in an ideation and learning session with ChatGPT.

Van Dusseldorp is one among millions of professionals around the world using large language models (LLMs) as work tools. Now, this technology is set to reshape learning and knowledge management, from sharing of information and expertise inside massive corporations to ongoing professional learning and development for solo operators. And that’s all before we get to the vast implications for traditional learning and education.

It amounts to a shift for learning as significant as the advent of the internet in the late 1990s. A transformation in the way we store, interact with, and share knowledge. So how are professionals across industry and academia using LLMs right now? What role are these tools playing inside large organizations? And what does the future hold?

We all know that LLMs have been the subject of huge excitement over the last couple of years. And sure, some of that — from talk of AI doom to predictions of imminent superintelligence — bears the hallmarks of a tech hype cycle. But get beyond the headlines, and it’s no surprise LLMs are being taken up in earnest by many professionals.

LLMs are built on transformer models: A special kind of neural network capable of teaching themselves about the underlying patterns in sequential data. When trained on vast amounts of text, transformer models learn about the deep statistical relationships between words as they are commonly used in sentences. The result is an AI with an amazing linguistic competence, such that it can understand natural language inputs and in response generate text that is relevant, detailed, and apparently meaningful.

This makes LLMs — think GPT-4 or Meta’s Llama 2 — a near-uniquely flexible knowledge tool. One with access to huge reserves of information and capable of generating natural-sounding text responses of all kinds.

Copy drafting — emails, presentations, and reports — are a clear workplace use. But now, many professionals are building research and ongoing learning practises around these tools.

Henry Coutinho-Mason is a futurist and speaker. He recently trained an LLM on two of his own books about consumer trends and innovation. Now he uses the resulting app as a research tool:

“I’ll step to the app and ask questions such as, ‘What do you think of the new Humane AI pin? Explain the trends behind this innovation,’ and it will come back with a first cut of ideas and insights,” says Coutinho-Mason.

“My work is about inspiring professionals to see what’s coming next, and to think about what it means for them. That means learning about emerging technologies and innovations is a crucial part of what I do. My customized LLM really helps with that.

“What people need to understand is that for uses such as this, it isn’t about the LLM giving me ‘the answer’,” he continues. “Instead, it’s about it providing thought starters, or giving me an insight into the overall shape of an emerging technology space. Then I can refine that. In that way, the LLM really accelerates my learning and research.”

No wonder, then, that large organizations are also experimenting with LLMs as internal tools. And here, the opportunities to supercharge learning are even broader.

Large organizations face particular challenges when it comes to learning and knowledge management. Knowledge and more informal kinds of know-how are typically distributed across the organization and stored in myriad ways, from documents, to slide decks, to spreadsheets and beyond. For even an experienced employee, finding the right knowledge source or person can take hours, days, even weeks of searching.

Now, some organizations are developing LLMs as a transformative new way to approach those challenges. In August consulting group McKinsey announced Lilli, an LLM fine-tuned on proprietary content spanning over 100,000 documents. It’s intended to act as a new way for McKinsey staff to access the vast storehouse of industry-specific knowledge, data, and more accumulated by the group over decades.

“With Lilli, McKinsey consultants can use technology to leverage our entire body of knowledge and assets… This is the first of many use cases that will help us reshape our firm,” said Jacky Wright, McKinsey’s chief technology and platform officer.

Associate partner Adi Pradhan, meanwhile, is using Lilli as a learning tool: “I use Lilli to tutor myself on new topics and make connections between different areas on my projects,” he revealed. “It saves up to 20% of my time preparing for meetings. But more importantly, it improves the quality of my expertise and my contributions.”

McKinsey are far from alone in exploring this new frontier. Morgan Stanley have fine-tuned GPT-4 on proprietary investing, general business, and investment process knowledge to create a conversational AI that can assist their financial advisors. Financial giant Bloomberg have created Bloomberg GPT, an LLM trained on its proprietary financial data and available for use by its own people and some clients.

The end destination? Soon enough many employees will come to expect access to these kinds of AI-fueled conversational entities. Think a workplace assistant, guide, and learning companion, ready to help 24/7. It amounts to a revolution in the way knowledge is distributed and absorbed. These companions are set to become key learning tools for staff — and they’ll play a key role in the induction and training of new employees.

It’s not only in industry that LLMs will transform learning. We’re already seeing powerful impacts in formal education settings.

Speak to educators, and you’ll soon learn that some of those impacts are challenging. They’re confronting a new form of plagiarism, in which students hand in work entirely or predominantly created by an AI. Tools are emerging to allow for the detection of AI-generated text but, just as fast, students are finding ways around them.

That race will continue. But longer-term the rise of LLM-fueled plagiarism will push educators to find new forms of coursework and assessment: ‘Write an essay on this topic, and then deliver a 15-minute talk presenting the core arguments of the essay.’

Meanwhile, LLMs are still prone to make factual errors, or so-called ‘hallucinations’. They can’t be relied upon to the exclusion of all other sources of information. But hallucination rates are already being cut dramatically by technical improvement, and that process is set to continue.

Aaron Woodcock is a lecturer at the University of Reading’s International Study and Language Institute. He convenes course modules for teaching partners in China, who are teaching academic English to their STEM subject undergraduates. That means lots of lesson planning.

“Up until recently I’d be given one or two modules to convene. Now there are two of us doing seven modules, so there’s a lot more lesson planning to do,” he says. “Using an LLM has made me so much more productive. I’ll chat to the AI and give it some of my notes and thinking. And then we’ll go back and forth and design a lesson. Then I’ll ask the AI to write that lesson as a full plan, and I’ll work on it.

“It still takes me two hours to get a lesson plan I’m happy with. But using AI makes it so much easier to get started, and that just makes the whole process faster. And I’ve had good feedback on the lesson plans from the teachers who are using them.

“It’s made me realize that I’ve always found writing hard, and always put tasks that involve writing off until the last moment. AI changes all that so it’s really improved the way I work.”

Woodcock is also using an LLM to process and synthesize insights from large volumes of student feedback. Managing and responding to this kind of feedback is now a significant part of life for academics and teachers inside higher education.

“The AI is immediately spotting patterns in the feedback that it might have taken me several days to see,” he explains. “We were more able to draw action plans out of the student feedback, and then we spent months implementing those actions. It has had a huge benefit on the course and really helped me with reporting and feeding back to the people around me.’

And on the AI plagiarism challenge? “Yes, there are challenges around student use,” says Woodcock. “But for me this is an opportunity to move away from a reliance on essay writing and regurgitation of knowledge.”

The University of Reading and many other academic institutions are now encouraging educators to explore new forms of learning and assessment, including more face-to-face assessments of knowledge - and thinking.

“This technology isn’t going away. We want to encourage students to use AI but in the best ways,” says Woodcock. “LLMs can also encourage diversity and inclusion; they will empower those who aren‘t necessarily at home with written language but still have so much to offer. Sure, there will always be people trying to pass off work they didn’t write as their own. But in the end, that’s no different from students buying essays online, which happens already.”

Where is all this heading longer term? A generation of students will become accustomed to working with LLMs as research, learning, and drafting tools.

They’ll emerge into the world expecting the ability to do the same in the workplace. And that will only incentivize employers further to train their own, bespoke AI models, offering them as 24/7 knowledge and learning companions for their people.

The future belongs to organizations — and individual professionals — best able to combine their own intelligence and creativity with AI in order to learn more, see further, and produce even better results.

We humans are unique in our ability to develop, store, and disseminate knowledge. That truth underpins our amazing capacity to cooperate at scale, which in turn gives rise to everything we build: Innovations, companies, universities, nations, and everything in between.

Our societies have always been shaped and reshaped by the knowledge tools we use, from writing itself, to the printing press, to the personal computer.

The LLM, and machine intelligence more broadly, is sure to bring transformations of its own — and we’re only at the beginning of the journey There’s still so much left to do; and much left to learn.

It's clear that the application, implementations, and future implications of LLMs are ushering in a new are for productivity in the future of work. To explore other captivating advancements that will influence the way we work, delve into our discussion with David Mattin on emerging work tech mega trends.