If you believe AI can’t replace programmers, you are deluded. “Oh, but it can’t code shit.” That’s completely wrong. It’s not 2022 anymore.
The code generated by LLMs is not as good as it is perfect in most cases. AI’s churning out code that’s as structured and reliable as what most people can produce. It often beats humans because it’s cleaner, more efficient, and error-free. It makes fewer errors than human programmers.
Programming is the ideal use case for machine-learned Local Language Models. And it will only improve as now, literally every month brings new updates and pushes whatever we thought wasn’t previously possible. It will turn the job market upside down. Entry-level coders won’t be valued by companies in the future. It’s already happening.
ChatGPT is such an effective help in programming that it has killed Stack Overflow. It scraped the code first, though.
Google reports now that 25% of its code is written by the AI, and Sundar Pichai says it’s “just the beginning.” Companies everywhere are secretly deploying AI at scale, even if they’re not keen to admit it (especially when it comes to the workforce). The small but noisy neo-Luddite circles screech at any mention of Artificial Intelligence, and admitting to the use of AI in reducing employment is as much a no-no for the public face of a company as not having a diversity hire policy.
That being said, I’m a big believer in the AI. It’s going to help us build great things, drive productivity into the sky, and push industries forward in ways we can barely imagine. Generative Artificial Intelligence is a breakthrough. It’s fantastic.
But until that happens, there will be shifts in the job market, and it will be dramatic. People who won’t adjust will stay behind.
I’d liken it to the automotive revolution.
In 1900, there were just 8,000 cars on the streets of the USA. There were not many paved roads, no infrastructure, no gasoline stations. People were still reliant on horses and trains.
But just ten years later, in 1910, the number of automobiles surged to 458,000 cars, thanks to a technical breakthrough of assembly-line production invented by Ford in 1908.
By 1920, the number of cars grew to 8.1 million.
We’re in the Ford moment now. Adapt and grab the opportunities generative AI offers, or you will become a stable boy (a junior) or carriage mechanic (senior) who lost their job, because no one wanted to ride a smelly horse anymore.
Milk that AI cow while it’s fresh. And do it quickly, because the competition in the job market is heavy and opportunity won’t wait around. The time is ticking.
One thing the AI does to the job market, is to force the employees to be more active. No more coasting along. If you sit back and wait for instructions, you are setting yourself up for failure. AI will demand initiative. You will have to be ambitious now. But it will reward you handsomely if you push forward and learn.
It’s the end of a certain era.
Here’s what you should do to do to grab that opportunity.
A robot firing the man from his job. Model: Flux.
Learn to Prompt
ChatGPT doesn’t generate flawed code. If it does, then it’s a “you” problem. Prompt engineering is the new skill gap. Good prompts are especially important in programming.
Don’t expect that ChatGPT will pull the code out of your head. AI doesn’t automatically “fill in the blanks.” You have to give it actionable instructions. Vague instructions lead to poor results. Ultimately, great prompting depends on your project and what you actually need to do in your work. What you get out is only as good as what you put in.
“Write a Python program to make a calculator.” is not a good prompt, because the AI is guessing about operations, interface, advanced features, or error-checking.
A good prompt should be more complex and would look like this:
Write a Python program that functions as a simple text-based calculator. The program should:
Display a menu allowing the user to select basic operations: addition, subtraction, multiplication, and division.
Prompt the user to input two numbers for each calculation.
Handle errors gracefully, including invalid input (e.g., non-numeric values) and division by zero.
Show the result of each calculation and then return to the main menu.
Keep running until the user chooses an option to exit the program.
Spend a bit of time learning to prompt in your job. Build your own prompt catalog.
There are plenty of resources to learn prompting and even more research papers. How to write effective prompts for large language models by Zhicheng Lin says you should do this to get the results:
- Add relevant context,
- Be explicit in your instructions,
- Ask for lots of options,
- Assign characters (roles),
- Show examples, don’t just tell,
- Declare preferred response format,
- Experiment.
It even includes the infamous example of “I will tip $200 for great responses,” which can increase the quality of responses. It has been probably patched up at this point anyway, so no point in trying it.
Learn to prompt GPTs – it will be a lifelong skill. It will be always relevant from now on.
Don’t Stop at ChatGPT
There are plenty of other models you should try out. You’re doing yourself a disservice if you don’t try them. Claude offers similar capabilities as ChatGPT 4o and o1 at coding. Some people argue it even outperforms OpenAI’s models in specific coding tasks. Both OpenAI and Anthropic GPTs produce clean, working code when prompted properly. Don’t sleep on Claude.
HuggingFace Arena Leaderboard for LLMs. Take note of Chinese Yi Lightning model. There are big developments in this field in China, and their models are now almost as capable as what OpenAI can bring.
That said, there’s a range of other models to consider for coding assistance, particularly open-source models. While the difference in capability often comes down to model size (smaller models usually have smaller context windows and less complexity, they are helpful tools.
OpenSource LLMs are handy for several reasons:
They are not censored: at least some models. ChatGPT will refuse to generate many types of code. It will even lecture you on what’s ethical and what’s not. That’s very arrogant. No one should tell you how to program.
A polymorphic virus written in Python and generated through Mixtral 8x7B. You can’t pull this off with o1 (it’s even more censored than 4o) and Claude. With open-source models, you decide what to do with technology. No one will lecture you what’s ethical.
They can be run locally: Smaller models, such as LLaMA or Vicuna, can even run on devices as lightweight as a Raspberry Pi or smartphone, and you don’t even rely on an external server or GPU. They can produce code, too, although nothing complex.
They are free: Both Claude and ChatGPT are paywalled. You don’t have to pay anything to run local models, except maybe cloud fees if you’re running them on a cloud.
They are private: Privacy is non-existent in the case of closed-source, cloud-based models. Your data is logged. You have no control over how your data is stored and used. It complicates many things at work. Companies are even introducing policies around ChatGPT usage, restricting employees from sharing passwords, proprietary code, or business details due to data privacy concerns. Local LLMs run on your machine, and they don’t have these limitations.
Get a Degree
There was talk just a few years ago about how the degree is useless now, as there is a lack of skilled programmers on the job market. But now everyone and their dog studies computer science. If you have no formal education, you are cooked.
Do you know how bad it looks like? Computer science degrees awarded are on the verge of surpassing ALL humanities degrees combined.
In 2022, 28.2% of degrees awarded were CS majors at MIT, which produced many headaches for the university, such as a lack of professors or a limited number of classrooms. It’s ironic that computer science faculty can’t overcome administrative obstacles.
The degree will be now more important than ever. Online bootcamps, and completing coding tutorials on Youtube will not get you taken seriously. You can have all the GitHub projects you want, but when it comes down to it, a degree will provide you credibility in a way few other things do. Having a degree shows you’re serious, and that you can grind. And forces you to actually understand why things work: algorithms, data structures, computational theory – boring or not.
The college might also plug you (potentially) in a network of people who can open doors, professors or alumni. And college life is great. It will teach you a lot about how to deal with roommates, build relationships, meet new people (a lot), and live on your own. I met my wife there some twenty years ago.
Finally, even if you’re self-taught, you’ll only make it to a certain level before they hit a ceiling. You won’t get past lower managerial jobs without a degree, let alone an executive position. A degree is the smartest, long-term investment you can make, and not only for your career. Don’t let anyone dissuade you from getting an education. They are suckers, they don’t know what they’re talking about, and they’re jealous of your future achievements.
Work Your Way Up
It’s tempting to think you can sit back and let your skills or resume carry you. But job security isn’t what it used to be. You’ve got to work your way up and keep moving. Don’t be satisfied with your current role. Always deliver, add value, and solve problems.
When you make yourself essential, you will be harder to replace by the algorithm. AI might be able to complete repetitive tasks (for now), but it won’t replace someone who knows how to keep things running. Everyone likes people who know how to keep things running smoothly. Take high-visibility projects, too. It will give you exposure to upper management and maybe let you use that leet code you have been grinding during six rounds interviews for that intern role.
You will have to build your soft skills, too. AI can crunch numbers and can replace analysts, big data researchers and similar positions. You can’t automate teamwork or emotional Intelligence (yet) and if you want to make yourself indispensable, you’ll have to show you’re a cool guy. Also, keep track of your achievements, metrics, and contributions to the company. Show it when you have to justify your position and make a case for promotion.
Patton, that overrated WWII general (I’m in a Matthew Ridgway team), in his speech to the Third Army, said:
“An ounce of sweat will save a gallon of blood.”
Patton said that in all of his talks, he “stressed fighting and killing.” Pour out that ounce of blood. Fight and kill in the corporate. You will save your pain later. And eventually you will get your promotion – with the job security that comes with it.
Patton was such a sigma (is this a correct word?). Although overrated.
In the same speech, Patton also said:
“We’re not just going to shoot the bastards, we’re going to rip out their living goddamned guts and use them to grease the treads of our tanks. We’re going to murder those lousy Hun cocksuckers by the bushel-fucking-basket.”
Inspiring! I don’t understand why we have to read Sun Tzu instead of Patton in business classes.
Stop Making Bullshit Excuses
The hanging of the Luddites at the York Castle, 1813. Contemporary woodcut print.
People are giving themselves plenty of excuses and even hate AI, to the point they treat showing disdain for Artificial Intelligence as virtue signaling. Don’t get swept by this wave. Be smarter.
“It’s just a tool” is what people say when they’re too comfortable in the old ways and want to ignore the new ones. Don’t ignore the AI. Sure, AI is “just a tool”- that’s changing entire industries.
But if you start learning how to use it efficiently, it won’t replace you, but it will push you to a new level. If you dismiss it as “just a tool,” you will be left bitter – and without a job.
Others say that the AI “is no substitute for real skills.” You could say that a year ago. But it’s not the reality anymore. AI is actively replacing people. Tasks that once required skills are now done faster, cheaper, and around the clock by AI. Programmers are in a good place anyway, but the customer service and data entry roles have it much worse.
Companies are already getting rid of those people where it makes financial and operational sense. Swedish payment provider Klarna has already cut its workforce from 5,000 to 3,800 in the past year and wants to reduce that to 2,000 employees by using AI in marketing and customer service. Klarna’s CEO said that the job cuts would mean Klarna could pay its remaining workers more. That’s why you have to upskill in the AI. Dedicating even just 15 minutes a day to learn new tools or work on your prompts will keep you advancing.
But the AI “lacks creativity!” Actually, Artificial Intelligence is more creative than you.
But the AI is “a capitalist tool made to steal our hard work.” Turn off your iPhone, comrade, and go outside.
The job market won’t make room for excuses. Quit rationalizing why you can’t or won’t do something, and introduce the AI into your pipeline. Otherwise, it will devour you. Artificial Intelligence is great, it’s empowering, and it will bring you opportunities – but you have to grab them.
Embrace the Technowizard in You
An illustration how Stable Diffusion works from my recently published book. Can you imagine there is a LoRA for generating Wojaks (Wojak SDXL)?
No one really knows how AI system works, especially more convoluted systems such as Stable Diffusion. Don’t let yourself tell otherwise. Neural networks were trained on massive datasets, and the results of machine learning are too complicated for us to understand. They operate with billions of parameters, trillions even. The idea of multidimensional tensors or a latent space is too complex for our brains and perception to comprehend.
Maybe you can solve a stochastic equation with the Euler-Maruyama method, and pat yourself on the back for that. But you will never be able to solve billions, trillions even of floating-point operations needed to generate a ChatGPT query. A single ChatGPT4 token can involve hundreds of millions of FLOPs. Good luck wrapping your head around that.
We are nothing compared to relentless GPUs that solve these equations. We are too dumb. We are too slow. We are outdated. AI’s logic operates on levels of abstraction and interaction too deep for us. We’ll never be able to comprehend what the AI knows and understands.
Euler-Maruyama method (incorrectly called Euler equation)
Roman Yampolskiy, in his 2019 paper Unexplainability and Incomprehensibility of Artificial Intelligence, says that AI decisions will never be explainable, even in principle, due to their complexity. Incomprehensibility means that even if AI systems could explain their decisions fully, humans will not be able to comprehend those explanations due to our intellectual limits.
Yampolskiy is the biggest AI doomer. He believes there is a 99% chance AI will destroy us eventually. First, there will be some kind of technofeudalism with the ruling caste of people bearing the keys to LLMs (all the more reasons you should grind your skills with Artificial Intelligence), and then the superintelligent machines will bring us down. Regardless if you think it’s true or not, for now, you should ride this wave while it lasts.
Ultimately, artificial intelligence models will be unexplainable. They already are. They are black-box systems where you can provide some input and maybe understand the algorithms involved, but that’s it. You will never be able to go toe-to-toe with AI.
That’s why you should become a technowizard and embrace these limitations.
You should learn everything that is needed to make the AI work for you. Try to understand transformers, token embeddings, attention mechanisms, and neural network operations if you’re working with LLMs or use programming or latent space concepts for graphical generative AI. Every new information piece of information you absorb, will make you better at guiding the AI to produce the code you need.
Embrace these tools, study what matters, and accept that the black box will always have mysteries. Wrangle the data, engineer your prompts, deploy some LLMs yourself, think with mathematics, and play with computer vision. Just don’t get that beard. It only looks good on scholars of post-Soviet origin.
Eventually, you will transcend. This level of understanding will set you apart from your colleagues. You will be able to turn the AI project to boost revenue, slash costs, or save money. This is how you win in this job market obsessed with ROIs. Start now.
