It was published Sept 2024.

The paper titled "How Large Language Models Can Reshape Collective Intelligence," published in Nature Human Behaviour, explores how large language models (LLMs) like GPT can transform collective intelligence (CI) in groups, organizations, and societies. Collective intelligence is the ability of groups to make decisions and solve problems more effectively than individuals. This is often achieved through mechanisms like crowdsourcing, prediction markets, and forums, where diverse inputs are aggregated to reach superior conclusions.

Key benefits identified for LLMs in CI include:

1. *Enhanced Collaboration*: LLMs can facilitate more inclusive and accessible online collaborations by translating languages, summarizing discussions, and potentially even representing individuals in deliberations.

2. *Accelerated Idea Generation*: LLMs can produce a large volume of ideas quickly, serving as both an ideation tool and a resource for brainstorming, particularly beneficial for less experienced participants.

3. *Support in Deliberation*: LLMs could assist in structured dialogues by guiding participants with questions and managing discussion flows, which could lead to more effective and inclusive decision-making.

4. *Efficient Information Aggregation*: By synthesizing opinions and identifying areas of consensus, LLMs can bridge gaps between individuals with diverse views.

However, the authors caution about potential risks:

- *Disincentivization of Human Contribution*: Widespread reliance on LLMs could reduce engagement in open knowledge commons like Wikipedia and discourage individual contributions to public knowledge.

- *Illusions of Consensus*: If certain perspectives are underrepresented in LLM training, the models might create a false sense of consensus, obscuring minority opinions.

- *Reduced Diversity*: Heavy reliance on LLMs might homogenize thought, reducing the diversity critical for effective problem-solving.

- *Propagation of Misinformation*: LLMs’ tendency to “hallucinate” or generate incorrect information could enable the spread of false narratives, especially in disinformation campaigns.

The authors recommend fostering “truly open” LLMs with transparent datasets, expanding computational resources for independent research, and instituting third-party oversight to monitor LLM usage. They argue that these measures can help balance the benefits and risks of LLMs, promoting CI while safeguarding against potential harms【5†source】.