The post Yet another idea of collaboratively edited knowledge base proposes to build a database of credible knowledge, understandable both to people and computers, including AI. This database could be used as a source of information in computer applications, and to increase the collective intelligence of people by easing and speeding up the exchange of knowledge and information between them. It can also be used to improve AI.
The current AI, including LLMs, has among others the following problems:
- the sources of information available for training LLMs often have low credibility;
- LLMs do not have detailed knowledge about the world; e.g., GPT will not tell you where to buy coffee nearby;
- LLMs often hallucinate;
- there is a risk associated with the creation of artificial general intelligence because large neural networks are black boxes and we do not know what they do internally.
The importance of the first three topics for the development of AI was discussed at the following times: 0:57:10-0:59:10, 1:00:30-1:00:50, 1:25:00-1:28:25 during the Twitter Spaces session about xAI in July 2023.
The proposed database of credible knowledge can be used with LLMs in two ways. First, it can be used in the training process. Second, it can be used with the trained LLM as an external source of credible information on any topic, if the LLM lacks it. Finally, because the knowledge database is human-readable, it is safer than a black-box LLM. The knowledge database can also be used as a source of knowledge for AI types other than neural networks.
In summary, the proposed knowledge database can enhance the intelligence of people and computers, making the AGI safer at the same time. Additionally, increasing the collective intelligence of people may also be an alternative method for achieving superintelligence.