My undergraduate degree was related to intelligent science, and I have some understanding of AI algorithms. Later, when I participated in an off-campus internship, I went to a professor's laboratory and participated in model training and coding for the first time. That was in 2012, the year when convolutional neural networks began to explode.
The recent ChatGPT is undoubtedly another event that refreshes human cognition with neural networks. After using this model for a while, I found that it still has some issues.
ChatGPT uses supervised learning and reinforcement learning, using a large amount of data sets to train the neural network. Artificially generated data is annotated for the AI, and model parameters are adjusted. In other words, ChatGPT will output answers that humans want as much as possible.
Previously, someone tried to make ChatGPT simulate a Linux host and output results for input commands. I don't think it actually runs these commands itself, but its model outputs the results that this command is likely to produce, which will likely satisfy humans.
Because I adhere to the principle of "laziness is the first productivity," I am one of the earlier users in the department to use ChatGPT to complete work (privately). I have tried using it to write documents, making modifications based on the content it provides, and completing a document that looks decent. Of course, I am the kind of person who uses it quietly, until another colleague loudly promotes how amazing this thing is in the office, then I know that my lazy path has come to an end.
The company's leaders are also trying to replace some people's work with ChatGPT, but in my opinion, this wish is currently not achievable. AI can assist humans in completing certain specific tasks to a certain extent, but for general tasks, current AI still falls short. We can use ChatGPT to complete some small programming tasks or text editing tasks, but the results it outputs still require some degree of manual editing. It is still not capable of achieving a "plug and play" level.
OpenAI is also an ideal company, wanting to solve all problems with one model, and ChatGPT is such a model (to a certain extent). I don't know what it will develop into in the future, but fortunately, we have the opportunity to try it.
In my mind, future AI will need a revolutionary creation to truly enter people's lives and complete various more complex tasks. Just like the resurgence brought to AI development by the burst of computing power in recent years, perhaps the next revolution will require better algorithms rather than relying on current neural networks (which may already be the best algorithms). No one can say for sure what will happen in the future. Perhaps in my lifetime, we might really wait for the day when algorithms are updated. Who knows, we might actually see a war between machines and humans (hopefully not).