Reprint the official account | Knowledge management is in Xiabo
A generative AI can only answer questions or complete tasks based on what it has been trained to do—unless it acquires external knowledge. The "current technique" for providing external knowledge is to embed and use vector databases, but anyone who understands Semantic Knowledge Graphs knows that it will give better results.
GPT-4 has been considered the "spark of general artificial intelligence" for a while, so I guess it's time to ask which one it prefers. The following is a conversation with GPT-4.
Question 1: You have two kinds of external information to choose from, one is a semantic knowledge graph and the other is a vector database. Think of the steps to complete a problem or task, which of these two storage types is more likely to give you the text you need as an LLM?
GPT-4: As an LLM, I would choose a semantic knowledge graph to complete a problem or task. Semantic knowledge graphs represent the relationships between entities and concepts in a more structured way, making it easier to identify relevant information for a specific problem or task. On the other hand, vector databases represent information as numerical vectors in a high-dimensional space, which may be more difficult to directly interpret and extract the desired text. The structured nature of the Semantic Knowledge Graph will allow me to easily navigate and find the information I need to complete a given task.
Question 2: Please give five examples of problems or tasks that explain why you made this choice.
GPT-4:
1. "What are the main causes of air pollution?" The semantic knowledge graph will establish structured relationships between entities and concepts, making it easier to identify the causes of air pollution. For example, it could have nodes for "Air Pollution" and "Causes," and edges connecting different causes, such as "Industrial Emissions," "Vehicle Emissions," and "Chemical Fuel Combustion."
2. "List the works of Shakespeare." In the semantic knowledge graph, there might be a node representing "William Shakespeare" connected to nodes representing his personal works, such as "Romeo and Juliet", "Hamlet" and "Mike white". This structure makes it simple to extract the required information.
3. "Describe the process of photosynthesis." The semantic knowledge graph will have nodes and edges representing the process of photosynthesis, including concepts such as "light", "chlorophyll", "carbon dioxide", "water" and "oxygen". A structured relationship between these concepts will make the process of understanding and description easier.
4. "What are the symptoms of the common cold?" The semantic knowledge graph will have a node representing "common cold", connected to nodes representing various symptoms of the common cold, such as "sneeze", "cough", "sore throat" and "runny nose". This structured information makes it easy to extract and list symptoms.
5. "Which countries are members of the European Union?" In the Semantic Knowledge Graph, there will be a node representing "European Union" connected to nodes representing its member states, such as "Germany", "France", "Italy" and "Spain". ". The connections between these nodes will make it easy to identify and list countries that belong to the European Union.
Problem 3: Hallucinations are a serious problem for LLMs like you. Do you think this problem can be eliminated if the information needed to perform the task is available in the Semantic Knowledge Graph?
GPT-4: While using a semantic knowledge graph can significantly reduce the chance of hallucinations by providing structured and accurate information, it may not completely eliminate the problem. LLMs, like any other machine learning model, remain error-prone and may occasionally produce incorrect or irrelevant information. However, the structured nature of the semantic knowledge graph undoubtedly helps alleviate the hallucination problem, as it provides LLM with more accurate and reliable information.
This is - what came out of GPT-4's mouth. Large-scale language model LLM needs access to external knowledge and data, if they are generally useful, semantic knowledge graph will be the preferred knowledge structure for LLM. If you are a developer or provider of semantic knowledge graph applications, there may never be a better opportunity like this again.
As answered by GPT-4, by effectively combining LLM and knowledge graphs, we believe that the power of 1+1>2 can be exerted, which is also the future of information research and knowledge management—we will start from traditional search/documentation Centered approaches turn to knowledge computing methods driven by large-scale language models and knowledge graphs.
OpenKG
OpenKG (Chinese Open Knowledge Graph) aims to promote the openness, interconnection and crowdsourcing of knowledge graph data with Chinese as the core, and promote the open source and open source of knowledge graph algorithms, tools and platforms.
Click to read the original text and enter the OpenKG website.