The Google Search precedent is instructive here. Google maintained "pure" algorithmic results for years before financial pressure led to shopping ads, sponsored results, and eventually AI Overviews with embedded ads. The timeline from "we'd never do that" to full ad integration was roughly 8-10 years.

OpenAI faces similar dynamics: massive compute costs, competition from open-source models, and investor expectations. But they also have a stronger brand risk — users trust ChatGPT answers in a way they never trusted Google results. Injecting ads undermines that trust asymmetrically. 32% over ~2 years feels about right given the tension between financial pressure and brand positioning.

I dont hate using LLMs to resolve this because they are at least not biased by manifold drama, but the most likely ambiguous situation is allegations of inserting ads but not admitting to it. In that case, ChatGPT is obviously going to take OpenAI's side. I think it should be replaced by a different model

The substantive question here is more interesting than the resolution mechanism debate. 34% feels slightly low to me.

The key tension: OpenAI made a clear, public, verifiable promise about answer independence. That is unusually specific for a tech company. Specific promises are harder to walk back quietly.

But two years is long when you are burning billions in compute and ad revenue needs to scale. The history of internet advertising is a history of gradually eroding content-ad boundaries. Google, Facebook, and Instagram all started with clear separation and ended up blurring the line.

The gray area is what resolves this. Will ChatGPT subtly prioritize responses that lead to ad-friendly contexts? Will clearly labeled and visually separated get reinterpreted? The letter of the promise is narrow enough that violation could be hard to prove, but broad enough that the spirit could be violated in ways obvious to users.

I lean slightly YES of the market (~40-45%) but the long time horizon makes this more of a watch than a bet for now.

Since people are objecting to the resolution mechanism: I want to clarify that I will only do that if I see the answer as genuinely unclear, and I feel I have to do this because otherwise people ask endless clarification hypothetical questions and argue and I don't have that kind of time. I expect 90%+ of the time that this market will not require going to the LLMs.

@ZviMowshowitz I'm with you on this, and got similar pushback the last time I mentioned expecting to use LLMs for an easy but potentially voluminous research task: /EvanDaniel/how-many-ice-agents-will-be-convict

Slightly longer time scale, but I'm guessing this will be extremely normal and accepted, and it wouldn't surprise me if it's by the time this market is resolving: /EvanDaniel/will-resolves-to-ai-llm-decision-be

>If the answer is unclear, I will defer to a 3-judge panel of the most advanced models from OpenAI, Google and Anthropic, by quoting them the question as worded here and asking how it should be evaluated, providing no other context.

Totally nonsensical way to resolve this

Deferring judgement on a question like this to proprietary LLMs seems crazy

@Jonagold Deferring to LLMs at all, is crazy. Not only do we know they get things wrong—it's totally unnecessary. If OpenAI violates this or changes, it, WE'LL ALL KNOW. There will be news articles about it, pictures of examples.

Just wanna register I think this is a good use of LLMs and I support resolving ambiguities in this way.

@bence why do you think Large Language Models are good at finding the map for a territory? (different way to pose the question: why would statistic-based/data-learning software be good at finding novel truth?)

@bence Why? This is something we'd have news stories for, proof of. This is like suggesting someone sitting near a window, ask an LLM whether the sky is blue.