Wikipedia already has very detailed guidelines on how text on Wikipedia should look, which address many of these problems.[1] For example, take a look at its advice on "puffery"[2]:

"Peacock example:

Bob Dylan is the defining figure of the 1960s counterculture and a brilliant songwriter.

Just the facts:

Dylan was included in Time's 100: The Most Important People of the Century, in which he was called "master poet, caustic social critic and intrepid, guiding spirit of the counterculture generation". By the mid-1970s, his songs had been covered by hundreds of other artists."

[1]: https://en.wikipedia.org/wiki/Wikipedia:Manual_of_Style

[2]: https://en.wikipedia.org/wiki/Wikipedia:Manual_of_Style/Word...

A good place to see this pre-2022 (the ai epoch) is articles on less known bands from the late 2000s when Wikipedia was becoming more popular. Quite a few of them turn out to be copy/paste promo text. I know this because I did webdev work for that industry, and when I look up those bands on wikipedia I will recognize the text as text that I personally had to paste into a bio page 20 years ago. Since the bands are well known, nobody reports it (I admit I'm too lazy)

The real tell on those tends to be weirdly time-specific claims that tend to be wildly outdated ("currently touring with XYZ")

fiction is part of our shared language and culture. we communicate by making analogies, and our stories, especially our old ones, provide a rich basis to draw upon. neither a person nor an llm can be fluent users of human language without spending time learning from both fiction and non-fiction.

What do you think the LLMs were trained on? 90% of everything is crap, and they trained on everything.

I'm thinking quite a bit about this at the moment in the context of foundational models and their inherent (?) regression to the mean.

Recently there has been a big push into geospatial foundation models (e.g. Google AlphaEarth, IBM Terramind, Clay).

These take in vast amounts of satellite data and with the usual Autoencoder architecture try and build embedding spaces which contain meaningful semantic features.

The issue at the moment is that in the benchmark suites (https://github.com/VMarsocci/pangaea-bench), only a few of these foundation models have recently started to surpass the basic U-Net in some of the tasks.

There's also an observation by one of the authors of the Major-TOM model, which also provides satellite input data to train models, that the scale rule does not seem to hold for geospatial foundation models, in that more data does not seem to result in better models.

My (completely unsupported) theory on why that is, is that unlike writing or coding, in satellite data you are often looking for the needle in the haystack. You do not want what has been done thousands of times before and was proven to work. Segmenting out forests and water? Sure, easy. These models have seen millions of examples of forests and water. But most often we are interested in things that are much, much rarer. Flooding, Wildfire, Earthquakes, Landslides, Destroyed buildings, new Airstrips in the Amazon, etc. etc.. But as I see it, the currently used frameworks do not support that very well.

But I'd be curious how others see this, who might be more knowledgeable in the area.

[dead]

"This is the smoking gun!"

_sigh_ Is it though, Claude, is it really?

I'm pretty sure Mistral is doing fine tuning for their enterprise clients. OpenAI and Anthropic are probably not?

I'm more thinking about startups for fine-tuning.

https://github.com/blader/humanizer

I didn’t know about that, thank you!

Yes, I did.

Usually they are smaller details in the pages, not the core claims, but that doesn't really refute my point of GPT being easily able to point them. Here are some examples, I'm not including all errors per page that GPT 5.1 found back then (this reply is already too long), but just a few examples.

https://en.wikipedia.org/wiki/Large_Hadron_Collider

> (infobox) Maximum luminosity 1×10^34/(cm2⋅s)

This is from the original design, LHC has been upgraded several times, e.g. if you check https://home.web.cern.ch/news/news/accelerators/lhc-report-r..., you see "Thanks to these improvements, the instantaneous luminosity record was smashed, reaching 2.06 x 10^34 cm^(-2) s^(-1), twice the nominal value." and that was in 2017.

> The first collisions were achieved in 2010 at an energy of 3.5 tera-electronvolts (TeV) per beam

This is wrong, if you check https://home.web.cern.ch/resources/faqs/facts-and-figures-ab... it says "23 November 2009: LHC first collisions (see press release)" - https://home.web.cern.ch/news/press-release/cern/two-circula... and the energy was 450 GeV

Another random example, I was reading https://en.wikipedia.org/wiki/Camponotus_japonicus (a very small article) and decided to ask GPT about it. It checked a lot of other sources and found out that no other source claims that this species of ant inhabits Iran.

Another one: https://en.wikipedia.org/wiki/Java_(software_platform)

> and—until its discontinuation in JDK 9—a browser plug-in

In reality it was deprecated in JDK 9 and removed in JDK 11 - most people would think "discontinuation" means that it was already removed in JDK 9

https://en.wikipedia.org/wiki/Nekopara

> The Opening theme for After, "Contrail" was composed by "Motokyio" and Sung by "Ceul".

Just two misspellings, it should be Motokiyo and Ceui

> A manga adaptation illustrated by Tam-U is currently being published

This section hasn't been updated, but the manga has already finished a long time ago.

===

Here's a direct part of GPT 5.1's response (I tried this back in November, so there was no GPT 5.2 yet) regarding luminosity, and it did also have a citation in the 2nd paragraph to the exact link I used above for the luminosity claim.

– The infobox lists “Maximum luminosity 1×10^34/(cm²·s)” without qualification.

– That number is the original design (nominal) peak luminosity for the LHC, but the machine has substantially exceeded it in routine operation: CERN operations reports show peak instantaneous luminosities of about 1.6×10^34 cm⁻²·s⁻¹ in 2016 and ≈2.0–2.1×10^34 cm⁻²·s⁻¹ in 2017–2018, roughly a factor of two above the nominal design.

– Since the same infobox uses the current maximum beam energy (6.8 TeV per beam) rather than the 7 TeV design value, presenting 1×10^34 cm⁻²·s⁻¹ as “Maximum luminosity” is misleading/outdated if read as the machine’s achieved maximum. It should either be labelled explicitly as “design luminosity” (with a note that higher values have been reached) or the numerical value should be updated to reflect the achieved peak.

> As a technique though, never ask an LLM to find errors.

What I do is both ask it to explain why there are no errors at all and why there tons of errors. Then I use my natural intelligence to reason about the different claims.

It gives references that you can then verify manually. I wasn't advocating for a 100% automated process.

Great way to get banned :)

I've made a bunch of nontrivial changes (+- 1000s of characters), none of them seems to have been reverted, never asked for permission, I just went ahead and did it. Maybe the topics I care about are so non-controversial no one actually seen it?

I agree. Wikipedia Cleanup is a good starting point. Or look for a Wiki Project to join.

I've found the best way to learn and contribute is to jump into an existing project. Usually direction is the hardest thing .

You can of course dive into an article and make changes, but you'll often get pushback (warranted or unwarranted) and that can be discouraging. It's a somewhat natural feedback loop.

Why shouldn't it be relevant?

The example you put forth supporting your claim that AI is perfect for Wikipedia editing is that you (ambiguously) edited an article, perhaps using AI.

The post also reads like it was partially written with AI.

I'm sorry you see my response as hostile, but I hope you can see how this example isn't accomplishing your intended rhetorical goals.

And also entirely ironic.

> Repetition and redundancy, ambiguity, illogical ordering of content, rambling sentences, opaque grammar.

Repetition AND redundancy?

Illogical ordering in an unordered list of things.

Rambling (could be more items or less).

What's the grammar of a sentence like this? Diagramming it would be a challenge. I'd call that opaque.