Don't Get Laid Off by AI (From a Lead Principal Engineer)

AI Job Displacement Concerns00:00

Lots of people are worried that AI is getting insanely smart, can change everything, and can take their six-figure tech job away. They are wondering what kind of skills they can build to position themselves well in that new world.

Your worry is 100% justified. Generative AI is a major breakthrough that's going to change everything. Some people say that it's at the same level as the invention of the smartphone, but that comparison misses the point. This is not just about making a computer smaller. Generative AI needs to be looked at as the discovery of fire or electricity. It can fundamentally change how we operate as a society.

If you are still doing exactly what you have been doing in the past, you will get left behind in that past, and by the time you realize what's happening it might be too late. You absolutely need to understand what is changing and how to plan for it.

How Large Language Models Work01:01

By now you probably have heard of the term ChatGPT or large language models. The quick intro of how they actually work: first we created this giant neural network which mimics the structure and complexity of the human brain. Let's think of it as an empty brain. Then we let this brain read the entire internet over and over again until it absorbs everything, and that means everything. Every book, every comment, every cat meme, every GitHub code file, everything. All of it gets absorbed into this empty brain.

Now we can ask the brain questions or have it complete tasks for us. We do so by providing context and the desired output as a prompt to the model. That's the basic form of large language models.

Very soon we discovered that it has a lot of limitations. To begin with, it's been taught everything up until 2021, so the brain hasn't seen anything that happened after that. As a result, we cannot ask this brain anything that requires recent knowledge. Teaching this brain is super expensive. It costs millions of dollars each time, so we cannot just refresh it every minute.

Solving LLM Limitations02:31

What did we do? We gave this brain the ability to read the internet in real time. So every time we ask the brain a question or have it do a task for us, it can first browse the internet for relevant knowledge and then use that to inform the output that solves the freshness problem.

Then we also discovered that the brain somehow sucks at maths. We probably just didn't wire it right. So what do we do to someone that's bad at maths? We give them a calculator. We gave this brain more and more tools like that in the form of plugins and functions.

There are also a lot of specific jobs such as teachers, lawyers, doctors, or therapists that the brain by default was not good at. Don't worry, we can fine-tune the brain with specific data sets to make it good at those jobs. This fine-tuning process can produce a lot of highly specialized brains, each being good at one specific job.

Context Window Limitations03:30

Another limitation is something called context window size. Each time we ask the brain a question or give it a task, we give that in the form of a prompt, and there is a limit for how long that prompt can be. Typically we can only give it a few thousand words, so we can't just ask the brain to summarize a thousand-page book.

Each time we talk to the brain, it will treat us as a new person it's never talked with before. That's also because of this context window size limit. It cannot remember that much. The solution to the problem is called a vector DB, which serves as long-term memory that the brain can use to store or retrieve information.

Jobs at Risk from AI04:01

The reason we spend so much time talking about how large language models work is because if that diagram is what your job looks like, then you are done. If your job is for someone on the other side of the computer to give you context and the task, let's say a Jira ticket or a product spec, and what you do is that you read the internet (say Google or Stack Overflow or even ChatGPT, you name it) and then you use a bunch of tools like Visual Studio Code, Excel, or Python to produce the desired output, then you're done. That's exactly what large language models are good at. AI is going to do your job thousands of times faster than you, and unlike you and me, AI does not complain about work-life balance.

Skills That Benefit from AI05:01

The right question to ask is what skills allow us to benefit from AI rather than being replaced by it. There are a few theories to share.

First and foremost, the ability to acquire context is going to be crucial. Remember, large language models work by accepting a prompt that has the context and the task. When someone has gathered all the relevant context and gave that to AI, AI can do the job very well. However, AI is not very good at acquiring context by themselves.

As an example, if you have a code base and you tell the AI the exact code file and function that needs to be optimized, AI can typically do that pretty well, pretty fast. However, if you just throw a problem statement like "our app feels slow," you can go get your popcorn and watch AI suck. That's because in order to solve this problem, you would first need to acquire some additional context on what it means for the app to be slow. This is typically done by real usage of the app or doing some performance profiling that can reveal which function is being the bottleneck.

By first knowing how to acquire this additional context, you can then summarize the context in the prompt and give that to AI to solve the problem. People who are able to acquire context and resolve ambiguity are much better positioned to benefit from AI than those who like to wait for others to give them context and bring all the clarity and they simply execute a task.

Mindset Shift Required06:32

To get good at acquiring context, we need to shift away from a task-driven mindset to a problem-driven, business-driven one. If your current job is to accept context and task from others, then you should be alarmed and start to shift your mindset ASAP.

Specialization Risk07:01

Aside from the ability to acquire context, the more specialized your job is, the more risky you are. How would you describe your job? If your answer is something like "I'm a React developer" or "I'm an Excel data analyst," then you are not in a good position.

It's very, very easy to fine-tune an AI model to be very good at one specific task. You can just give it a lot of examples and it learns very fast. There are literally thousands of startup companies who are doing that fine-tuning as we speak, and their business model is literally to replace humans with AI at a lower cost and they split the difference.

It is, however, much, much harder to build general AI models that's good at everything. So people who are versatile and can connect dots between multiple domains will be a lot harder to replace.

Versatility as Defense07:31

If you are an engineer, try to not lock yourself into a specific tech stack or language or framework. Expand your breadth and go full stack. Don't shy away from non-engineering tasks either. The more versatile your skills are, the more defensible your role will be in the new AI world.

You want to be the person who's commanding an army of AI models to do work for you, not someone who watches AI become better and better at their job.

Decline of Craftsmanship Value08:30

It makes me a little sad, but the value of craftsmanship will continue to decrease. The world will become increasingly in need of entrepreneurs who are outcome oriented.

Think of the Industrial Revolution. Who benefited more? The carpenters who continued to produce a few handmade chairs per month, or the ones who build factories that produce chairs at massive scale? Although machines cannot achieve the same level of craft as a skilled carpenter, it can get close enough while being able to make products orders of magnitude faster and cheaper.

The same goes for tech jobs now. AI will be able to write highly scalable, maintainable, and performant code a lot faster than humans ever could. That's not to say craftsmanship is no longer important, but if that's what you talk about all day and it's all that makes you proud, you need to be aware of the risks coming your way.

Ideally, you should expand your skills a little bit more towards the entrepreneur side. Start to understand business success. Start to take on business accountability. And start to shift your mindset towards a business outcome.