Large Language Models and the jobs they will enable/replace

When ChatGPT came out, it was THE fastest growing product of all time. Ever. Here is a table to give you some perspective:

It took only 5 days to reach a million users. Is this because it has to do with text and conversations? Maybe? Prior to this, Facebook actually released a version of their chatbot a few months prior but not very many people talked about it in comparison. It’s still around https://blenderbot.ai/. The difference? Context.

ChatGPT’s context is much larger than any of its rivals, which makes it more human like and conversational. What’s even more interesting is that you can sign up for Open AI’s API and give it any persona you like and it will perform accordingly. It’s like programming with text instead of code.

Jobs

This means it can take on any role that involves text and soon images (see GPT-4 and Mini-GPT). What all requires text? Hmm. Seemly everything done on computers that requires a keyboard.

• Code
• Copywriting
• Marketing
• Presentation Making
• Business Strategy
• Research
• Blogging
• Legal
• Some aspects of Medical

These are just those I could think of so I asked ChatGPT what jobs it could help out with and here is what it came up with (notice that it’s more through):

1. Content Creator
2. Social Media Manager
3. Digital Marketer
4. UX/UI Designer
5. Data Analyst
6. Project Manager
7. Customer Service Representative
8. Sales Representative
9. Virtual Assistant
10. Language Translator
11. Researcher
12. Technical Writer
13. Online Tutor
14. Graphic Designer
15. HR Recruiter
16. Business Consultant
17. Financial Analyst
18. Medical Professional
19. Legal Professional
20. Educator.
21. Editor
22. Social media manager
23. Marketing strategist
24. Business analyst
25. HR specialist

This is a large portion of the white-collar jobs out there. For now they won’t be replaced, but you can use these models to do your marketing for you, or to explain and write code. It can also explain topics of physics and neuroscience. It can even do creative things like explain string theory in the form of a rap.

Tools

People are building tools that can automate these tasks so eventually you will reach a point where some of these professions will be replaced with technology.

Here are some examples:

Github Copilot – Writes code for you

Jasper.AI/WriteSonic/WriteCreme – Writes content for you

Cocounsel – Legal assistant powered by GPT-4 which has aced the bar exams

No public tool yet but GPT-4 will assist doctors in diagnosis

Intercom’s GPT-4 Integration – Customer service powered by GPT-4

Limitations

It’s only a matter of time and GPU/CPU resources until those jobs are replaced. Currently it’s expensive to run these models because they require so much GPU capacity. It’s literally running layers of math functions over your input and using probabilities to determine the output of the next letter to write. The context is also very limited or expensive to run for larger contexts. To get around this, people are using vector databases to perform a semantic search against text/documents, then they stuff the context with the part of the relevant data. This appears to give the large language model enough context to appropriately answer questions but it’s not perfect. Some might argue that that’s what makes the large language models appealing because they make mistakes like humans.

Considerations

Because of the scope of the impact that this technology has to disrupt the workforce, there are considerations that need to be taken into account. What will those people that will get replaced by this tech do? Will they become prompt engineers? The person that can chat with the computer the best wins? Does everybody become a chat engineer? Or have we reached Moore’s law and it becomes impossible to make these more efficient, making them virtually impossible to make profitable resulting in nothing to worry about?

Conclusion

As large language models continue to evolve, their potential to replace jobs that involve text is becoming more apparent. While it may not happen immediately, the rise of tools that automate tasks once performed by humans is inevitable. However, the limitations of expensive GPU/CPU resources may slow down this process. These models are not quite perfect yet. As we continue to witness the evolution of language models, it is important to consider the impact they will have on the workforce and begin preparing for the future.

Note that there are new models that can interact with other models, such as those that can generate images or convert images into text. This may increase the impacts of this technology.

Examples:

HuggingGPT– A GPT model that can interact with any other AI model.

Mini-GPT4 – A GPT model that takes images as input, converts it into a description which then can become context for regular large language models. Yes, you can draw a website on paper and it will convert it into functional code for you. You only need a developer to create the server and put the code there.