oh another one is LLM. does it count reasoning models, agentic models, etc.

@Bayesian is Nano Banana an LLM

@Bayesian I'll take a crack at this:

A "large language model" or "LLM" refers to any machine learning system that (1) is trained primarily on text data (natural or/and programming languages) using self-supervised learning objectives such as next-token prediction or masked language modeling, (2) contains at least ten million learned parameters, and (3) is capable of generating natural language or programming language text.

This definition includes transformer-based models, state-space models (such as Mamba or its derivatives), recurrent architectures, hybrid architectures combining multiple approaches, and any future architectures that satisfy the above criteria.

This definition explicitly includes the following, provided they meet the core criteria: reasoning models that incorporate extended inference-time computation, chain-of-thought processing, or deliberative reasoning steps; instruction-tuned or chat-optimized models; models fine-tuned with reinforcement learning from human feedback (RLHF) or similar alignment techniques; agentic systems where an LLM serves as the core decision-making component, even when integrated with external tools, code execution environments, web browsing capabilities, or multi-step planning frameworks; and multimodal models that accept image, audio, or video inputs alongside text, provided text generation remains a primary output modality.

This definition explicitly excludes the following: pure image generation models that do not generate natural language as a primary function; speech recognition or text-to-speech systems that lack generative language capabilities; traditional NLP systems based on rules, templates, or non-neural statistical methods; embedding models or encoders that produce vector representations but do not generate text; and models under ten million learned parameters, regardless of architecture.

For resolution purposes, parameter counts should be determined by official disclosures from the model developer when available. If official parameter counts are unavailable or disputed, credible third-party estimates from peer-reviewed publications or established AI research organizations may be used. The parameter threshold applies to the total count of learned parameters, not to active parameters in sparse architectures such as mixture-of-experts.

I think the acronym ORN ("Otherwise, Resolves NO") should be more widely used since it's such a common phrase.

Queer

@jim rude! @Bayesian is just trying to help!

If timezone is not specified and cannot be inferred from market closing time, assume Pacific Time.

@Robincvgr how do you know whether or not it can be inferred from market closing time

Recommend not using terms "Pacific Standard Time", "Central Standard Time", "Eastern Standard Time" etc to refer to times during which daylight saving time is observed (and vice versa). Instead, use daylight saving time to refer to times during which it is in effect and standard time to refer to times during which it is not, or use the relative "Pacific Time", "Central Time", etc.

@Robincvgr I am going to continue using "Manifold Time" and everyone else will need to just live with it.

@Eliza I fully accept any repercussions.

@Eliza Lol I was just about to say something about that

67: a term with no meaning

reputable news source

@Thomas42 plus extra fun stuff like what if it was 'hacked' or rogue employee or otherwise retracted, etc. etc. etc.

I don't want to need to spell out every edge case every time.

@Thomas42 This might help, even allows for laxer standards if wished, by specifying which checkmark or better is required. https://en.wikipedia.org/wiki/Wikipedia:Reliable_sources/Perennial_sources

Wikipedia:Reliable sources/Perennial sources - Wikipedia

@JussiVilleHeiskanen @Bayesian @Thomas42 I support "green checkmark on Wikipedia's perennial sources list at time of individual market creation" as a definition for reputable news source

@Robincvgr TBF, that list may be ones they have history about discussing, not an exhaustive list of reputables. So it is biased to those questioned, that is edge cases. So I don't support it as definition, but rather as an additive term, like a list of reputable sources, AND green mark on perennial wikipedia sources ALSO acceptable. But those ARE scrutinized and thus screened.

AGI

@jim any recommendations?

@Bayesian a totally unserious term whose use should result in the market being unranked and treated as a meme market

@jim based ok i'll add

What/who is a furry?

"individuals who like to dress up as animals"? but that's not exactly verifiable, maybe there's better? preferably something that is in most cases publicly verifiable

@Bayesian is self identification necessary or sufficient?

@Eliza stating self-identification would be sufficient i think. not necessary bc some ppl that show many straightforward and public signs of being furries may have not verbalized it and it would be dumb for a market to resolve unintuitively bc of that

@Bayesian what about the furries that are just in it for the porn?

@NivlacM hmmm "or attracted to other individuals because of the fact that they dress up as animals"? idrk

Win a professional cycling race or stage: Defaults to the winner declared on the day at the podium celebration after the race. Later disqualification or alteration of the results (the same day, next day, or any time in the future) would not change the resolution.

I'm thinking a single term can have multiple definitions depending on the context? like announce (of a company) has to be official or wtv but announce (of a person) is just if they say the thing, even in an unofficial context? stuff like that

in other words your def is good and might be best placed under the category of "win" in the subsection "of a professional cycling race or stage" or wtv