Original | Text by BFT Robot
In 2023, with the explosion of GPT in all walks of life, "whether GPT can be used in scientific research scenarios" has become a natural question. When ChatGPT surpassed most humans and achieved staggeringly high scores in fields such as college entrance examinations, SATs, U.S. law exams, and medical exams, people's interest in GPT-driven scientific research grew. As of the release time of this report (August 2023), LLM’s practice in the field of scientific research has begun to emerge, and the application ecology has become increasingly rich. Throughout the history of mankind, science has attracted us with its mysterious charm, constantly leading us to explore the unknown world, answer the mysteries of life, and promote the progress of society and the development of civilization.
Now, while artificial intelligence technology is changing people's daily lives, it has also begun to affect human exploration of the edges of science. We are at an intersection where artificial intelligence and scientific exploration are beginning to collide and merge. At the center of this intersection, Large Language Models (LLMs) are serving as a new medium to reshape the way we interact with knowledge itself. In this article, we will systematically explore the transformative potential and challenges of LLMs in scientific research. First, we will elaborate on the view of LLMs as a two-way interface between human beings and knowledge; then, we will deeply explore the boundaries of LLM's ability to handle scientific content and improvement methods; finally, we will have an open discussion on the strongest potential of LLM at present - the thinking chain. (Chain of Thoughts), and its possibility to revolutionize human scientific research methods.
The possibility of AI as an interactive interface between humans and knowledge
Printing deserves its place among the four great inventions of mankind. Everyone's life is limited. However, the knowledge created by each person in his limited life is distilled and precipitated, recorded in text, passed down from generation to generation, and updated iteratively. This paradigm can be said to be the foundation for the survival of human civilization.
From primitive people carving oracle bones to today, people can call complex digital databases at their fingertips. Human beings are always pursuing more efficient methods of acquiring, storing and disseminating knowledge. Precisely because humans use language as the “solvent” of knowledge, large language models (LLMs) like GPT-4 have the potential to revolutionize the way we interact with scientific knowledge. Central to this revolution is the concept of LLM as a two-way interface, a dynamic medium through which people can extract and contribute knowledge.
This dual function redefines humanity's relationship with the boundless scientific literature. This is particularly important today when disciplines are constantly intersecting and knowledge is constantly converging. The progress and development of science is undoubtedly driving our society forward, but in this process, a major contradiction becomes more and more obvious: that is, the contradiction between the increasingly vast amount of knowledge and the limited capacity of the human brain. Various fields of science are intertwining and intersecting at an unprecedented rate. Like the collision and fusion of galaxies in the vast universe, these intersections and collisions have produced entirely new areas and far-reaching impacts.
The most obvious example is the "bioorthogonal reaction" that won the 2022 Nobel Prize, which combines the design and optimization of chemical reactions with the complexity and particularity of biological systems to emit selective, efficient and biocompatible reactions. reaction system to meet the needs of biological research, medical diagnosis and treatment. Another example is bioinformatics. This emerging field is the combination of computer science and biology. It is through the calculation and analysis of biological data that we can have a deeper understanding of genes, cells and ecosystems. For example, "ab initio molecular dynamics" has changed microscale scientific research, which is the fusion of electronic structure research and molecular dynamics simulation; another example is quantum computing, which perfectly combines quantum physics and computer science and has the potential to completely change our Understanding and application of computing.
This trend also brings a challenge: scientific researchers need to understand an increasingly wider range of knowledge, and acquiring, understanding and applying this cross-field knowledge is undoubtedly an arduous task. It is against this background that the emergence of large language models (LLM) brings us hope.
On the one hand, LLMs make the extraction and synthesis of knowledge efficient and convenient. By deciphering and presenting complex scientific information, they significantly lower the barrier to entry into new fields. This ensures that even novices can navigate and understand scientific discourse, even if they may be intimidated by the complexity of professional literature. For example, ChatPDF allows users to upload a PDF file (usually a scientific paper) and then ask an AI agent questions about the paper, and the AI can understand the content of the user's PDF and answer it based on its own knowledge - just Just like a professor preaching, teaching, and answering questions to his students. Such an ability is undoubtedly liberating for scientific researchers. They no longer need to waste a lot of time processing irrelevant information, but can directly focus on their own research areas or explore new research areas. This enables researchers to access and enter new fields faster and better achieve cross-field research.
In the future, when the barriers to entry into science are significantly lowered, not only young people who have just entered science will benefit greatly; scientists who have been working in professional fields for decades will also be able to easily overcome increasingly subdivided subject barriers. Find a stone that can attack jade in other mountains. In the world's leading universities, interdisciplinary informal exchange activities are often held to promote the collision of knowledge and wisdom in different fields. The emergence of LLM can make such a mechanism work asynchronously on a global scale breaking the limitations of time and space, tearing down the walls of disciplines and allowing knowledge to flow freely again.
On the other hand, LLMs can speed up and improve the process of knowledge contribution. For example, using LLMs' ability to perform multi-step reasoning and decision-making, researchers can find the most relevant papers in the maze-like breadth of the scientific literature. This not only speeds up the literature review process but also improves the quality of scientific papers by ensuring a comprehensive and relevant review of existing knowledge. This process is not a simple information retrieval, but a deep learning and multi-step reasoning process. It is this kind of reasoning ability that can help researchers find the most valuable knowledge in complex information.
In addition, LLMs can also play a key role in writing scientific papers. By assisting in language production, they can provide essential assistance to scholars, especially those for whom English is a second language. LLMs not only provide linguistic assistance but also help construct logical narratives and ensure coherence, which is crucial in scientific discourse as complex ideas and findings must be carefully organized and presented.
However, LLM means much more than that. On a larger scale, LLM gives more people the opportunity to contribute knowledge. Knowledge is no longer an ivory tower within high walls and is no longer the exclusive preserve of a few. Anyone, regardless of their language, background or status, can participate in knowledge exploration and innovation through LLM. This is undoubtedly a huge progress and a necessary tool for human civilization to enter the next stage. Without such tools, knowledge development may only be monopolized by a few traditional institutions, leading to stagnation in innovation.
Today, with the rapid development of science and technology, the old methods of knowledge acquisition and innovation can no longer satisfy our desire for progress. The emergence of deep learning and artificial intelligence, as well as the great success of institutions such as DeepMind and OpenAI, have proven to us that new scientific research methods can promote the development of knowledge more efficiently and innovatively. Why were world-shaking achievements like AlphaFold and ChatGPT (and the Transformer behind them) not born in universities? This soul torture may be worthy of consideration by university administrators. These new scientific research methods are not meant to replace universities, but they hope to allow more people to participate in knowledge exploration and innovation in a more open and fair way.
AI will not replace universities, but universities that embrace AI will surpass those that resist AI.
Humanity's thirst for knowledge is endless. The emergence of LLM not only solves the problem of acquiring and processing cross-field knowledge for us, but also makes the intersection and collision of science become freer and smoother. We have reason to expect that this change will bring us more scientific breakthroughs and innovations, which is indeed exciting and exciting.
Author | Chunhua
Typography | Spring Flowers
Review | Cat
If you have any questions about the content of this article, please contact us and we will respond promptly. For more information, please pay attention to BFT intelligent robot system~