I. #

Part of the reason I decided to study maths was because it seemed like the most useful subject. If I knew ML systems would never get better than GPT-2 at maths, I’d probably be of the same opinion today. But today’s state-of-the-art ML systems are far better than GPT-2. LLMs have excelled at maths and programming because maths- and coding-related tasks admit quick feedback, allowing for efficient reinforcement learning. Jobs involving applying maths and programming jobs could theoretically be automated within a few decades. The glory days of the nerd might be over soon.

I’m not going to speculate in the details of how such a transformation might occur; I couldn’t be bothered doing the forecasting due diligence here. Instead, I want to focus on future-proof jobs and skills.

There’s no shortage of material on this topic¹, so I’ll start by re-hashing some common beliefs. This article is a good starting point.

Current AI systems are bad at complex physical tasks and long-horizon tasks, and rapid improvements in any of these areas seem unlikely. So, we might expect human plumbers and wedding planners to be around for some time². There’s also a broad class of jobs all about meaningful human interactions. These jobs might be difficult to automate by virtue of the machine being a machine. For example, consider priests. Some people argue that healthcare workers also fall into this category of jobs, but I think might could change; for example, people are already getting more used to the idea of LLM therapists.

There’s also widespread agreement that leadership skills and personal effectiveness will become increasingly important. As will creativity, the ability to learn quickly and to use to AIs.

II. #

So many skills, so little time! But I’ll argue there’s one skill that’s vastly more important than those mentioned above: verbal communication skills. By this, I mean the ability to deliver good talks as well as the ability to articulate your ideas clearly in debates and normal conversations. A good conversationalist might have a huge comparative advantage in an AI-dominated economy.

To a large extent, LLMs can help with the other skills mentioned above. A manager can use LLMs to sanity check their decisions, to write better project proposals and even to manage intrapersonal relations. As for personal effectiveness, one of the five buttons in the Claude menu is “Life stuff”; clicking it displays five prompts related to getting things done. It’s also obvious that AIs make us more efficient learners. Finally, AIs can help you use AIs more efficiently: just have another LLM refine your prompt.

How about verbal communication skills?

LLMs can help writing scripts for talks. But that’s about it. The LLM can’t help you deliver the talk. This involves adopting the right pace, using the right intonation, reading off the audience, and so on. Good orators are rare! In spontaneous conversations or debates, you’re basically on your own. You can use an LLM to refine your ideas in advance, but you can only prepare so much.

People who can communicate clearly in speech seem much more credible, at least to me³. Suppose I write a persuasive article but fail to explain its contents to you. Then you might feel somewhat disappointed, thinking that all the good points of the article were all due to an LLM. In fact, had I written a mediocre article, I might have appeared more credible overall.

Even if we ignore the impact of AI on the job market, the ability to express one’s ideas clearly in speech is essential, regardless of what job you have. Becoming a better speaker also has positive spill-over effects. You’ll certainly have more good conversations with the people you care about.

Even if we tomorrow develop AI capable of automating all remote work, integrating these AIs into the workforce would take time – just try imagining the amount of bureaucracy! For this reason, I’m not in a rush to develop verbal communication skills. Rather than doing rhetoric MOOCs and buying books on persuation with cringe titles, how about just having coffee with people with shared interests more often?