Should LLM generated market description text be clearly labeled?

Mar 31

Yes

It's fine how it is

For example, the text below, which was generated by clicking this button:

I'll search for relevant information about LLM-generated content labeling practices and any existing standards or discussions around this topic.#### Resolution criteria

This market resolves to a poll voted on by other users. The question asks for a normative judgment about whether LLM-generated market descriptions should be clearly labeled, which is inherently subjective and depends on individual values regarding transparency, trust, and disclosure practices.

Background

The creation of content with LLM tools must be fully disclosed, and platforms like Medium require stories incorporating AI assistance to be clearly labeled as such. Under the EU AI Act, LLM providers are now required to clearly label AI-generated content. Multiple U.S. states including California, Florida, and New York have passed laws mandating disclosures when content includes AI-generated material. China's "Measures for Labeling AI-Generated Content" (effective September 2025) compel all online services that create or distribute AI-generated content to clearly label such content.

Considerations

Research shows that AI disclosure labels may have only weak or no effect on people's perceptions and use of AI-labeled content, suggesting that labeling policies serve functions beyond persuasiveness, such as helping users make informed decisions. Academic frameworks distinguish between mandatory, optional, and unnecessary disclosure of AI use, arguing that disclosure should be mandatory only when AI use is intentional and substantial.

Market context

Get

1,000

to start trading!

People are also trading

By 2027, will it be generally agreed upon that LLM produced text > human text for training LLMs?

62% chance

By 2029 end, will it be generally agreed upon that LLM produced text/code > human text/code for training LLMs?

77% chance

Will LLMs' loss function achieve the level of entropy of human text by the end of 2030?

61% chance

Illegal Agent-like LLM which automatically serves up links to copyrighted texts available by mid 2026

29% chance

At the beginning of 2028, will LLMs still make egregious common-sensical errors?

67% chance

LLM or AI written US Federal Laws before 2027

29% chance

LLM-generated comments on Manifold Markets Stock (Permament)

Ṁ269

4 Comments

Sort by:

I think any communication that could be assumed to be human-to-human but in reality uses AI should be labeled.

@ImaPerson I agree. I think it's a legitimately tough question, with fuzzy borders that cover "raw LLM" to "lightly edited LLM" to "LLM, but with careful conversation and back and forth" to "I talked with an LLM and then wrote it myself" to "no LLMs came anywhere near this". I'm not sure where I'd like to put the "this should acknowledge the LLM" boundary, but unedited text should be on the labeled side of it, in my opinion.

Yes because it is important to know if the human who made the market has thought about it for more than 5 seconds. If they have not, don't bet. You will be surprised how often the AI description is contradictory or the human makes "corrections" in the comment section that change the probability dramatically.

@Eliza exactly.

I've used LLMs to help write criteria before, but it's a process and involves review and editing and asking it about edge cases, not clicking this button.