Large Language Models and Dimensions of Productivity

In recent years, various high-intelligence technologies have emerged with the continuous development of artificial intelligence and machine learning. From the early rudimentary recommendation algorithms to the captivating AI-generated illustrations, and later, popular large-scale language models like ChatGPT, each technological innovation challenges our understanding of artificial intelligence. Nowadays, AI technology is constantly pushing the boundaries of our cognition. As with any generation-defining technology, some people are enthusiastic about it, while others resist it. Some people boldly accept it, while others fear and detest it. After reading others’ opinions on ChatGPT, I have also formed my own thoughts on this matter.

ChatGPT and Large Language Models

I believe that we should all have a consensus that, so far, ChatGPT or large language models led by ChatGPT may not be as important or impressive in and of themselves. Their importance lies in the validation of a bold idea. At this stage, ChatGPT may be able to address some simple and general domain knowledge, but for more in-depth issues, critical thinking is needed to examine their truthfulness. Additionally, ChatGPT currently has certain accuracy and correctness issues that require long-term correction and prompts to give us the answers we want. Therefore, we cannot completely trust what it says or be 100% sure that it will provide the most correct answer.

However, even though this is the current situation, what about the future? Technological development always proceeds at an exponential growth rate because the increase in production tools can provide positive feedback to the production of the tools themselves. Of course, although this acceleration is a trend, such development needs to be considered at a macroscopic level. However, after the success of ChatGPT, it’s hard not to think about how future large language models will impact us and what kind of impact they will have on productivity. What I am contemplating is the disruptive influence that future large language models may bring.

Dimensions of Productivity

What is productivity? If we must define this concept deliberately, its interpretation may seem monotonous. Therefore, I would like to explore the manifestation of productivity in different dimensions from my perspective.

Manual labour is the most basic and primitive dimension of productivity. This dimension may be the most basic and elementary expression of productivity. However, this aspect can be replaced by assembly line machines. Since the Industrial Revolution, people have gradually used energy and machines to replace simple and repetitive mechanical operations. The efficiency improvement brought about by this change is self-evident. Even though physical labour can be replaced, people at that time, or even earlier generations, believed that abstract concepts such as knowledge and content could not be “assembly-lined”. However, this is precisely the second dimension.

The second dimension of productivity is the language and knowledge level. In fact, this layer covers various forms such as knowledge divulgation, Q&A, and content creation, etc. In the future, large language/visual models will replace those mechanised knowledge carriers. Actually, I had thought before that today’s software engineers are still engaged in mechanised work to some extent. It’s just that they have changed from carpenters in the old days to carpenters in the new era. Perhaps one day, they will be replaced by some kind of automated carrier, but much earlier than I expected.

In summary, ChatGPT can proficiently use communication skills and become a modern encyclopedia. I believe that large language models can accurately explain the depth of professional domain knowledge and can be regarded as a knowledge baseline. ChatGPT can fill in any knowledge gaps in any aspects for users who use it until they reach the knowledge baseline-which is some basic, general, and generalised domain knowledge. But for more in-depth domain knowledge, further existing knowledge may be needed to evaluate its authenticity.

In the future, large language models will efficiently replace the previously cumbersome and redundant information output, carrying, integration, and re-output work. They will comprehensively replace positions that only deal with surface-level knowledge processing rather than just replacing knowledge carriers. This also means that the “learning power” of simply memorizing knowledge and the “abstraction power” of converting knowledge into transferable abstract concepts are no longer unique abilities of humans and are no longer productivity dimensions that people should strive to cultivate. One example is that in the education field, we should no longer emphasise memorizing knowledge but instead start thinking about the third dimension - “acquiring higher-level learning and abstraction capabilities”.

The Third Dimension

I recently watched a video about ChatGPT, which mentioned that we should learn higher-level learning and abstraction abilities. However, I have a different perspective. Indeed, I agree with the view that encourages people to develop creativity, higher-dimensional learning ability, and abstract understanding ability. But after watching that video, I noticed ChatGPT’s excellent abilities in learning and abstracting. It can learn the concept of “human learning ability” in the non-linear output of “emergence” and apply it to the model’s learning process. This phenomenon means that its abstract ability can be abstracted again on the basis of abstraction, exhibiting complex nesting and recursive structures. This makes me wonder if it is necessary for people to make efforts to develop higher-level learning and abstraction abilities. Can we surpass large language models? If we cannot surpass them, why bother investing in this dimension?

In terms of physical strength, the answer provided by the industrial revolution is that humans cannot surpass mechanical arms in pure physical power. In terms of calculation, computer technology has proved that we cannot surpass computers in pure computing power. What is underway now is the competition between our language which is carrier of human creation, communication, and memory; and natural language generation models, and the current situation seems unfavourable. Can our ultimate reliance on learning and abstraction (higher-dimensional understanding and creation) really lead the way for future super-large comprehensive models?

I believe that the third dimension should not only be higher-level learning and abstraction abilities. We need to find areas where humans can uniquely shine, areas that cannot be replaced by automation, and then enhance human productivity based on this foundation. But before that, it is difficult to determine what the third dimension of productivity that can be invested in will be in the future.

Human-Centred Approach

Despite concerns that AI will replace humans, the fact is that ChatGPT’s primary service target has always been humans. In fact, the worry that AI will achieve the same status as humans or even replace humans is unnecessary. Humans have a flaw or feature, which is desire and demand, which AI cannot mimic. Due to the different interests and needs between humans and AI, there is no direct conflict of interest. Therefore, we do not need to overly worry about AI taking away human jobs or replacing human positions. If we can continue to exert creativity and innovation in areas where humans have advantages, then we will be able to better utilise AI technology to improve productivity and bring more value to human society.

Due to the uniqueness of machine models, to some extent, the “thinking” of models will become closer and closer to biological thinking of humans. The direction I am researching is brain-computer interfaces, which attempt to digitise the human input-output process. To some extent, brain-computer interfaces help accelerate the similarity between AI models and human thinking abilities. As this field continues to develop, the ability boundaries between the two will inevitably become increasingly blurred.

In a human-centred world, many industries face the risk of being replaced by models that offer higher productivity. However, I believe that technologies related to human perception and perception are currently the only ones that cannot be replaced. For example, using machine learning models alone, it is difficult to give people a real sense of experience from activities such as riding a roller coaster. In addition, there are technologies that provide various ways for people to experience different experiences, such as fully immersive virtual reality technology. Through these technologies, people can engage in various activities in a virtual environment and obtain feelings and experiences similar to those in the real world. Therefore, I believe that brain-computer interface technology is also opening up possibilities in these areas.

Dealing with Uncertainty and Anxiety

In fact, everyone is afraid and anxious at some degree. This kind of anxiety is normal. We are afraid of unemployment, being replaced, new things, feeling useless, disrupting the peace, falling behind in learning, others being more familiar with tools, losing the meaning of our existence, and afraid that AI will bring not only potential but also negative consequences. But think about the pandemic, we all found our anchor in the midst of anxiety. In the future, we can only choose for ourselves. Accept or reject, the times will continue to move forward. And the anxious you, haven’t you been inspired by this magical thing with even a trace of childhood curiosity?

The demand to replace humans is ultimately just part of a balance between costs and benefits in a game. Just like there are still assembly line workers and farmers, these are some examples that are be replaced completely. I believe similar situations will also occur in various fields. In the end, no matter what happens, there will always be people who enjoy horseback riding, assembling machines, creating for the sake of creation, and those who enjoy activities that involve exploring knowledge in depth. Even if their expressions are not accurate enough, perfect, or lack creativity, they still create and express themselves.

Conclusion

After finishing the blog post, I sat in my chair and tried to slow down my fast-moving brain. Taking two deep breaths, I wanted to take a break and watch some anime. However, when I just opened the webpage, I suddenly realised that all future creations may be generated by artificial intelligence, and I felt a little sad. You ask me what my feelings are about artificial intelligence – whether I am happy or anxious, how should I answer?

Well, it’s a bit regrettable and somewhat tasteless. Once there were things that we thought were unique to humans, somewhat romantic, somewhere we were proud of, somehow sacred, I am afraid my friend, are no longer just ours.

Finally, this was written by the passionate ChatGPT =).