Does personal computer really run personal software?
We've been using personal computers for decades, but the software we use are mostly built by several software vendors. The problem is, the whole software industry has been dominated by a small group of people who have the ability to write code, a user has to be 'techie' just to open a terminal and start writing code. The recent emergence of LLM in the software industry will bring about a monumental transformation. LLM provides a vehicle for every software to use LLM as human intention understanding, also LLM provides a tool for every end user to write small piece of code without learning how to do traditional programming. Soon all normal computer users will have the ability to develop small software tools from scratch by describing what the software looks like and to describe modifications they’d like made to software they’re already using.
How exactly LLM can help?
Extract structure information from user's input
A simple example, let's say you want to build a simple appointment system, your system will extract the date and time from the user's input, and then call the Google Calendar API to create an appointment. Before LLM, it's a real hard problem even for hardcore programmers, the reason is there are numerous ways to describe a date and time, and the user's input can be very noisy. For example to describe next Wednesday, you can say "next Wednesday", "next week Wednesday", "Sep 7, 2023", "the day after tomorrow" and so on. It's a real hard problem to extract the date and time from the user's input. But with LLM, you can simply describe the date and time in natural language, and LLM will automatically extract the date and time from the user's input, and then you can call the Google Calendar API to create an appointment. The following is a simple example of how to use LLM to extract date and time from user's input.
Write small piece of code, with almost 100% accuracy
Although the basic idea of binary search is comparatively straightforward, the details can be surprisingly tricky
There are not so many people can write a binary search algorithm correctly, but with LLM, you can simply describe the algorithm in natural language, and LLM will automatically generate the code for you. With today's technology, LLM can generate code with almost 100% accuracy. Human is good at planning, but is not good at handing details and corner cases, but LLM is good at it.