Zombie stats: Where the numbers AI quotes come from | Hacker News Reader

Zombie stats: Where the numbers AI quotes come from

trakkr.ai

183 pointsmektrik74 comments

To me the dream of AI is: an agent can do the drudge work of tracking every statement down to the supporting reference, documentation and ultimately the work involved in verifying.

It is NOT being used for this, instead we want faster “answers” that avoid due diligence. Same as always but lazier.

I mean you can definitely do that - just set it up do to that.

What you are describing is that a majority of humans are inherently not interested in that. There are are a subset of people who would like to do that - and certainly can now that they have the tooling.

Its just a vastly strong tool. If you want it, build it.

I think it's more that the majority of humans DON'T REALIZE it's NOT doing that.

Most people think AI answers are carefully sourced, triple vetted, over-researched to death and can trust it with their life, despite the warnings, until they run into nonsense and then dismiss it as a one-off.

If there was a "always give the best possible, most researched and trusted answer" flag that was easily visible, or maybe "chatGpt" and "bestAnswersGpt", where the second one did all the verification and was considered much better overall, guess which one people would start using after a while and/or quote, because of improved accuracy?

It's not that people don't WANT honest information (well, most people) - they just don't KNOW they're not getting it right now. That's the problem with tech - they're trading on "computers were always right in the past" because there has been no generally available "fuzzy search" type application that needs handholding to give "good" results. Most software either gave the result of computations that are always the same, or enforced static rules (like spellcheck) which were perceived as "always right".

‘AI’ isn’t giving unsourced answers. Hastily cobbled together chatbots based on LLMs, inadequately prompted and without adequate research tools, are being put on the internet by humans and giving unsourced answers.

Well prompted LLMs with appropriate tools can do exactly what OP asked for. So AI can do this. That the AI-based products people are being given don’t is not a limitation of AI, it is a failure of humans.

Google overview frequently cites webpages that don't include the keywords of a query.

This is Google, the most used search engine in the world, made by a company that is at the forefront of AI development.

ChatGPT, Claude, etc,. constantly generate answers without sources and rarely cite anything verbatim. I'm not sure if they could cite something verbatim even if they wanted.

Essentially none of these tools are appropriate for information.

I've had the same thought as probably have many others - what solutions exist in this space?

To put it into words: a layer over an LLM that tracks provenance: the documents it has access to, the facts it has extracted, the mathematical expressions and queries it's run and finally the artifacts and results it's yielded complete with receipts for exactly how it arrived at it's findings and conclusions.

I assume legaltech products have to do it like this. Is the general pattern implemented somewhere on github?

Not sure why a solution is even needed. If I had to compile all the resources I had consulted for something like C programming, they wouldn't amount to a single shelf. I would say the same thing for research where the amount of sources is still very low. It's kinda reminded me of the big paper manuals and books before the internet. I don't think the whole library of congress was needed to write UNIX.

So AI being a source of truth and information is not a something that important. It's a generator and we should use it as such, getting it to map domains and repeating patterns. It should not be used as an encyclopedia or oracle.

>> If I had to compile all the resources I had consulted for something like C programming, they wouldn't amount to a single shelf.

Serious question - with something like a programming language where there is a sort of finite information available about the language and how to use it. I've found once you start moving into areas where this isn't the case like metaphysics where there is a lot more ambiguity, does the sourcing and information then become more valuable? Or are we going to finally be ok with ambiguity as humans?

I can see what you're saying and agree, but are there areas where sourcing does in fact become more important?

> does the sourcing and information then become more valuable? Or are we going to finally be ok with ambiguity as humans?

I think they do become more valuable. Who said something and who believe it is equally as important as what was said.

The name of the solution is RAG - retreival augumented generation. It is pretty easy to whip up your own over a corpus of material. I built one for myself within a week over a corpus of my favorite cook and homebrewing books. All you need is source material, postgress (pgvect) and some compute (llama.cpp with a decent chat model for chunking and labeling, an embedding and a rerank model. or some $ to use those via a router). When the corpus is indexed, you use your favorite coding agent and a system prompt that forces the model to answer only from search results.

I'm working on something around this that I'm hopeful I can open source in the next couple of weeks. It's just a research project, so not like a product that can be bought and shipped to prod right away, but I'm excited about it. I think a lot of people are working on this right now, at least in part, with things like memory layers and other approaches to "grounding", or with some of the more recent stuff around certification and governance. But there's sort of a split right now between:

- Groups focused strictly on model capability (i.e. benchmark scores)

- Groups really focused on governance, interpretability, control planes, etc

- A small but excitable minority of weirdos (me included) who think they're the same thing

A lot of the stuff that gets loosely clustered under neurosymbolic methods might be interesting if you're looking to dive in. The sort of pattern you're describing is something that a lot of people have taken stabs at from different directions, or at least have tangentially nipped at while working on related problems.

yeah i work on this at https://groundedai.company/

we build a product called Veracity focused on citation and claim checking. legal is a major focus for us, also scientific and technical publishing, where we detect hallucinated citations in submitted works and all that jazz

planning to release something soon focused on the LLM harness layer

we've been working on the problem for years. gradually getting traction

all completely closed source atm, though we often revisit that discussion

i would contest a sibling comment - the solution is not RAG, at least not entirely. 1) because it doesn't necessarily involve 'generation' (e.g our grounding model doesn't depend on gen ai) and 2) RAG is just enriching the context. even if you can get the retrieval right (not a given) it doesn't guarentee that the LLM will faithfully or correctly use that context, or that all its claims are supported by sources. RAG is not the same as traceability and verification

Agreed, and there is already a solution that’s better to support anyway. It may not be perfect, but Wikipedia operates on this model at scale at a much smaller annual budget than any of the AI companies and has produced much higher quality content overtime for all of humanities benefit than the AI teams which are most just scraping the work of others while spending most of their original work to replace human labor for their own financial benefit.

Wikipedia community bans AI-generated entries because verifying and removing AI hallucinations imposes even more workload on editors than writing articles from scratch [1, 2].

If you don’t agree that Wikipedia’s approach, one can always fork it like was done with SpaceX / Twitter did with its launch of Grokipedia on October 27, 2025, which hasn’t been really incorporating much improvement well with lots of digressions in its quality and reliability [3].

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

[2] https://en.wikipedia.org/wiki/Wikipedia:Editing_policy#Artif...

[3] https://en.wikipedia.org/wiki/Grokipedia#Reliability

[Edited for better formatting]

The problem with Wikipedia is that quality varies greatly across fields. Technology or software engineering are great, physics varies, but e.g. even articles about the most basic math concepts are completely incomprehensible.

I can often ask an AI to give me an easily understandable three paragraph overview of a concept that on wikipedia has a hundred paragraph article requiring multiple prerequisite textbooks to even start to comprehend. And that's for basic math concepts like multiplication or exponentiation.

But soon, the majority of those references, documentation and work will be AI generated. We are peeing and likely defecating in our own water well.

This is exactly how I use Claude in all of my technical writing. I even wrote a skill for this.

Every claim has to be backed up by a reference. Every reference has to be double-checked. Primary sources are much preferred over secondary sources, and where no primary source can be found, that must be noted.

I also insist on simpler, more easy-to-understand writing, even if that makes the total output longer. Aiming at a general audience, of the same reading level as a standard newspaper.

Most of the rest of the skill is about ensuring the Chicago Manual of Style is properly followed, widow/orphan control and how that applies to callout boxes, tables, etc….

link? care to share? :) i often find myself starting with a tone guide.

Not yet. I want to publish my skill on GitHub and then let other people contribute ideas to it.

But as it stands today, there’s still too much personalization that is currently built into that skill.

I need to remove that personalization stuff and make it more general. Once I’ve done that, I am planning on publishing it.

I can't decide if the report's writing ("One number, followed" & "Quoted by all five. Sourced by none.") is being lazy or ironic?

Probably researched and written entirely by an AI…

Google's AI is extremely prone to making up random shit that doesn't exist. Swedish is an agglutinative language (like German) so you can join multiple words together to form new words: tomat (tomato) + jos (juice) => tomatjos. Most conjoined words are nonsense, but Google will still happily "explain" what they mean. Some examples:

tangentbordskanin (keyboard bunny): keycaps for mechanical keyboards with bunny images on them

indianbokföring (Indian accounting): simplified accounting outside of legal norms

bokföringsindian (Indian accountant): informal, older term for an auditor

potatisbensinskämt (potatoe petrol joke): a type of classic absurd riddles or word plays

nagelfis (nail fart): bulging pain in fingers after having been outside in extreme cold without gloves

And it regularly lies about local stuff, e.g., it says that coffee shop A has lower prices than coffee shop B when it's the opposite in reality.

I know Perplexity isn't often talked about anymore but their design of every sentence needing to be able to be referenced back to an actual source (like an academic peer reviewed paper) seems like a super power for trust.

How do they ensure that?

No one ensures what it does but it’s proactive, it gets the people buying

Agents in Paris?

Blades of glory for the original quote.

I want it "that" way.

They fine tune for that, they don't ensure anything.

(Or at least they used to, I think they changed everything recently, but I've mostly stopped using them.)

My initial experience with Perplexity was it citing AI blog spam as a primary source so consistently I immediately dropped my subscription. (this was over a year ago, I haven't re-evaluated them since)

Is it up to Perplexity to determine what is AI blog spam & what is correct. Versus what exists online?

I'm not sure what the question means.

Whether Perplexity is useful to someone, and how useful, will depend on how effective it is in giving answers that are (1) correct as often as possible and (2) checkable if they're sometimes wrong.

If Perplexity's answers often treat AI-generated blogspam as reliable then that will probably make them less often correct (because people generating AI blogspam generally don't need what they generate to be consistently correct) and harder to check (because people generating AI blogspam generally don't need what they have to be clearly sourced, and because even if they did it's an extra step of link-following).

So if Perplexity turns up a lot of AI-generated blogspam, that probably makes it less useful. Is it "up to Perplexity"? Again, I don't know what that means, but I don't see that it matters. What matters is how much value it provides, at what cost in time and money and effort and attention. If its output is mostly repeating claims from AI-generated blogspam and that makes the answer "not much, and at too much", then so much the worse for Perplexity even if there's some sense in which it's under no obligation to do better. If its output is reliable and checkable, then so much the better for Perplexity even if some of its deficiencies are in some sense moral failures.

I’ve added to my system prompt in perplexity “ignore unreliable sources” and “make sure claims are supported by multiple sources” and that seems to help.

Of course for anything of consequence I’d go and actually read the sources (but for plenty of every day stuff it’s fine). On the free plan btw

> I’ve added to my system prompt in perplexity “ignore unreliable sources”

Ah, the good old magic trick!

It works surprisingly well sometimes (:

LLMs are really weird sometimes. I had some issues using a coding harness where the agent was failing to use the correct tool call syntax (I was using some rando free model on OpenRouter). I sent a message “look closelier at the tool call format” then all the tool calls for the rest of the session succeeded.

> I’ve added to my system prompt in perplexity “ignore unreliable sources” and “make sure claims are supported by multiple sources” and that seems to help.

Don't forget to add "don't make mistakes!"

They're referenced here and don't appear to score much better, assuming this page isn't itself a parody and just zombie stats themselves.

Yeah, until you follow the link only to find out that the link does not say what Perplexity told you it does. This happened to me a frightful number of times.

A full 40% of AI stats actually have a specific study behind them? Knowing how the human baseline is that 73.6%* of all statistics are made up on the spot, this is really encouraging.

*That particular stat was of course made up, but not by me.

Just an thinly veiled ad for something called trakkr. Not to mention that none of these "stats" actually matter aside from being the type of thing that a consultant would put in a pitch deck, at which point their accuracy doesn't matter anyways

The other major application is justifying opinions you already hold (and cited as a bludgeon against people who disagree).

And this doesn't even touch on the fact that a significant fraction of the published, peer-reviewed, authoritative sources ultimately turn out to be wrong[1]. Science is hard. Data collection and cleaning is hard. Mistakes are easy to make.

[1] https://journals.plos.org/plosmedicine/article?id=10.1371/jo...

Ssssssh, we are not supposed to talk about the massive reproducibility crisis.

Oh. Sorry, my bad.

What about the fraud? Can we talk about that? Or the Sokal-oid "we'll publish nonsense" journals?

Seriously though, we should be talking about this. The greatest engine for progress humanity has ever built is rotting from within and its animated corpse is being used as a status-granting rubber stamp machine by the same old priesthood of "trust me, bro, God said I was right" but now in lab-coats instead of robes.

I absolutely loved the grievance study affair lol. It’s all too funny, but it’s actually a genuine problem that could lead to the eventual downfall of the whole enterprise, if it’s a massive towel of babel build upon rotten wood

Never trust a statistic you haven't sneaked into the training corpus yourself.

Today I learned that "snuck" is a considered by some to be a non-standard American word, because I read that sentence and mentally puked, then did some research.

All this is to say, I fucking hate AI and enjoy research.

This article is complete slop and not worth looking at, but it's an issue worth discussing.

One of the original php2shell announcement posts (which was clearly AI written) claimed that 70% of websites use PHP. This number has now been repeated by every single article about php2shell, including in the most trusted tech news outlets.

I think it could have been approximately true in 2008, but not in 2026. The number is too round anyway. Trying to search for a source to the claim is now impossible.

If the basis of a reliable fact is consensus, then every AI model should now consider this fact to be true. In reality consensus is not useful of course, offhand facts are frequently generated by citogenesis. An article mentioning a stat offhand is not a source. A source should take some amount of epistemic responsibility for a claim.

Looking at HTTP Archive's September 2026 crawl[1], PHP shows up on 4.22M of 8.74M mobile origins, which is 48.3%.

My guess is the 70% figure comes from W3Techs[2], which currently puts PHP at 69.2%, but only of sites "whose server-side programming language we know." Given that PHP is uniquely trivial to fingerprint (via WordPress markup, .php paths, PHPSESSID cookies, etc), it is very likely that it is overrepresented in their dataset[3].

[1] https://httparchive.org/

[2] https://w3techs.com/technologies/overview/programming_langua...

[3] https://en.wikipedia.org/wiki/Streetlight_effect

PHP URLs don’t have to include .php, though?

PHP apps don’t have to use the session; if an app does not start a session deliberately, there’s no cookie. Some PHP frameworks don’t use the built in session. Bespoke PHP code may simply never use it. Or perhaps they do but they rename the cookie. Laravel doesn’t use it, for example. Magento does, but it renames the cookie. Prestashop stopped using it.

Web servers don’t have to announce they use PHP at all.

It is no more uniquely trivial to fingerprint than ASP is/was, for the same reasons.

There’s really a lot of dark PHP out there; I maintain a bespoke app that you would not be able to easily establish was PHP.

One reason why the 70% figure is worth going with is that you barely need any PHP on a site to exploit php2shell anyway. It’s a runtime vulnerability. If it’s possible to provoke the server to hand a URL to a PHP runtime, it’s potentially possible to exploit. A hell of a lot of cheap hosting has ready-to-go support for PHP.

That's interesting, good find. Those stats from W3Techs would be much more useful if they just admitted how much of their sample they are discarding! Maybe if somebody paid

They do also say they don't count subdomains separately, so wordpress.com is one site.

I think you are interpreting the data on httparchive wrong. 'All technologies' does say 8.74m, but again I think that is only successfully fingerprinted sites? If we filter to top 10m ranked sites, it goes down to 6.47m, with PHP at 3.19m (so 32%). Desktop is even lower for some reason.

There is a human/user element here as well - will the users check the citations, if they are provided inline? I recall in the early ChatGPT web search and Perplexity days, most of the citations didn't work (made up, or 404). You get a false sense of security by having a "[1]" next to the stat or a claim. Maybe labs companies should have some additional quality ranking/checks on these, if it's in their interest of course.

Well, 78% of all statistics humans mention are made up as well, so.. :)

Good findings, but the findings are so AI-generated and hard to read… Feels like I am being thrown numbers without a coherent story, and even worse, with beautiful animations which only serves as distraction… This looks very similar to the diary entries my LLM writes when working on a coding project.

The animations made it practically unreadable on my phone. Hope someday these people realise they still need to pay designers (they won't)

It just goes to show that when AIs are relied upon to recall information, they may hallucinate.

It lends credence to the theory that a smaller model without the ghost of information past, could do better with ability to search the web, although this is heavily reliant on the effectiveness of the search engine itself.

While it's interesting to see which models performed better I wish they did a control. Gave the question to x number of people and had them provide an answer.

bad (or even made up) statistics have been floating around meat space since forever.

The same thing happened to me while building quote extraction agents. AI would often extract figures that are not stated anywhere in the quote document. This has reduced with models getting better, but we still ask it to cite the exact line it got the figure from.

ive found that my claude with my preferences generally gives studies with real numbers cited. BUT if i go in incognito mode it gives me the worst citations imaginable. i once did this and was appalled at how bad the stats were. i worry for people who dont think to double check sources on numbers, or even just demand the ai double checks the validity of the numbers or gives links so the user can check the validity themself.

I don't trust LLMs to give me numbers like this, and it's not just that it might make a mistake on the number or it might have garbage in its context - it also might just not understand the mildest nuance in its context or apply any common sense.

I recently asked what percentage of ice cream is air. It told me up to 100%. When I followed the sources, as I always do and not just with silly answers like that, the number in its context was indeed up to 100%. There's an industry term called overrun, measured in percentages. But 100% overrun means it's 50% air.

So why would I trust any of these numbers in this obviously LLM-generated and probably LLM-researched article?

I guess it is slightly clever, if you feel like slipping out posts, to make them anti-AI. Really blunts the effectiveness of criticism of putting slop out when the slop agrees directionally with the criticism.

Ah yes. The AI slop blogpost calling out AI slop statistics. We must go deeper

If you want articles about AI to be written by experienced users of AI who have skin in the game, and criticize AI from the perspective of being active users receptive to improvement, you kind of have to accept that those people are highly likely to use AI to write articles.

The determination with which AI and fitness trackers deliver numbers, plots, and statistics is really seductive.

Better to be on the 40% or the 60%? Seems like both sides will bite you in the ass to an extent

So no different, than humans repeating facts we hear every day as 1 in 5 dentists tells us

There's no point to AI's existence if it's dumber and more careless than the baseline stupid human.

Clearly what they're going for when they called it OpenDenistSmalltalk

AI, turning the joke that half of all statistics are made up into an accurate statement.

An apple a day keeps the doctor away and Santa wears red.

There is a number of things one could do to reduce the share of made up info in the generated text, yet at the end of the day, to guarantee 100% factual accuracy of the informal prose output, a human has to go and manually check every single claim. No other way, just the nature of LLMs.

That's why a transparent attribution graph is paramount for trust-worthy AI. Otherwise paid disinformation campaigns, as already spewed out by for-hire private-spy agencies (NSO and similar) and 'LLM-SEO' in general, to poison LLMs will deteriorate LLM search to become as enshitiffied as the 'classic google/bing/etc ad-driven' search beforehand.

when people say "there should be no AI taught in schools ever. prompting? what skill is there in talking to a liar bot?" this is where they are dead wrong. Being correctly critical of LLM output is a huge skill and news flash it applies to people and human-written things too.

Being skeptical/critical isn't a skill, it's an attitude, a state of mind. An education that requires you to find your own information and sources manually breeds that state of mind. Giving LLMs to kids (or even adults) nurtures the opposite. It's like taking them to McDonald's every day and telling them to eat healthy. Yes there's a salad on the menu, but it's rarely going to be chosen.

Of course the reality that adults do trust AI uncritically because it's the path of least resistance means the traditional education I mentioned above wasn't sufficient. But that's all the more reason not to introduce a variable that makes things worse.

> Yes there's a salad on the menu

The salad is also pretty trash if you use the default dressing.

Being critical of LLM output isn't a new skill that is separate from critical thinking.

Being particularly critical of LLM output after using LLMs is not an acquired skill either; what is acquired is the erosion of trust.

Reply on news.ycombinator.com