How Much Do LLMs Hallucinate in Document Q&A Scenarios? A 172-Billion-Token Study Across Temperatures, Context Lengths, and Hardware Platforms [TLDR: 25%]

Comments

• RandAlThor@lemmy.ca ⁨2⁩ ⁨weeks⁩ ago

  This is pretty bonkers. How TF are they fabricating answers???

  source

  • bad1080@piefed.social ⁨2⁩ ⁨weeks⁩ ago

    https://en.wikipedia.org/wiki/Hallucination_(artificial_intelligence)

    source

    • snooggums@piefed.world ⁨2⁩ ⁨weeks⁩ ago

      Aka being wrong, but with a fancy name!

      When Cletus is wrong because he mixed up a dog and a cat when deacribing their behavior do we call it hallucinating? No.

      source

      • -> View More Comments
  • Zink@programming.dev ⁨2⁩ ⁨weeks⁩ ago

    I’m no expert and don’t care to become one, but I understand they generally trained these models on the entire public internet plus all the literature and research they could pirate.

    So I would expect the outputs of those models to not be some kind of magical correct description of the world, but instead to be roughly “this passes for something a person on the internet might write.”

    It does the thing it was designed to do pretty well. But then the sociopathic grifters tried to sell it to the world as a magic super-intelligence that actually knows things. And of course many small-time wannabe grifters ate it up.

    What LLMs do is get you a passable elaborate forum post replying to your question, written by an extremely confident internet rando. But it’s done at computer speed and global scale!

    source
  • ji59@hilariouschaos.com ⁨2⁩ ⁨weeks⁩ ago

    Because guessing correct answer has is more successful than saying nothing.

    source
• jacksilver@lemmy.world ⁨2⁩ ⁨weeks⁩ ago

  Just for context, this is the error rate when the right answer is provided to the LLM in a document. This means that even when the answer is being handed to the LLM they fail at the rates provided in the article/paper.

  Most people interacting with LLMs aren’t asking questions against documents, or the answer can not be directly inferred from the documents (asking the LLM to think about the materials in the documents).

  That means in most situations the error rate for the average user will be significantly higher.

  source

  • rekabis@lemmy.ca ⁨2⁩ ⁨weeks⁩ ago

    As I pointed out in another root comment, the average - depending on the model being tested - tends to sit between 60% and 80%.

    So this opens up an interesting option for users, in that hallucinations/inaccuracies can be controlled for and potentially reduced by as much as ⅔ simply by restricting the model to those documents/resources that the user is absolutely certain contains the correct answer.

    I mean, 25% is still stupidly high. In any prior era, even 2.5% would have been an unacceptable error rate for a business. But source-restriction seems to be a somewhat promising guardrail to use.

    source

    • jacksilver@lemmy.world ⁨2⁩ ⁨weeks⁩ ago

      Thanks for providing the actual numbers.

      I think one of the more concerning things is, what if you think the answer is in the documents you provided but they actually aren’t. What you think is a low error rate could actually be a high error rate.

      source
• FauxLiving@lemmy.world ⁨2⁩ ⁨weeks⁩ ago

  At 32K, the best model (GLM 4.5) fabricates 1.19% of answers

  Not bad, I don’t know many people who are 98.81% accurate in their statements.

  source

  • snooggums@piefed.world ⁨2⁩ ⁨weeks⁩ ago

    Calculators are correct 100% of the time.

    source

    • FauxLiving@lemmy.world ⁨2⁩ ⁨weeks⁩ ago

      Calculators are not people, Mr. <1.19%.

      source

      • -> View More Comments
  • Iconoclast@feddit.uk ⁨2⁩ ⁨weeks⁩ ago

    It’s a pleasure to meet you! The only thing exceeding my wisdom is my modesty.

    source

    • FauxLiving@lemmy.world ⁨2⁩ ⁨weeks⁩ ago

      Truly the most humble person of all time.

      source
  • Lemming6969@lemmy.world ⁨2⁩ ⁨weeks⁩ ago

    You can be wrong and not fabricate. This is closer to human intentional lying.

    source
• SuspciousCarrot78@lemmy.world ⁨2⁩ ⁨weeks⁩ ago

  Firstly, thanks for this paper. I read it this afternoon.

  Secondly, well, shit. I’m beavering away at a paper in what little spare time I have, looking at hallucination suppression in local LLM. I’ve been testing both the abliterated and base version of Qwen3-4B 2507 instruct, as they represent an excellent edge device llm per all benchmarks (also, because I am a GPU peasant and only have 4GB vram). I’ve come at it from a different angle but in the testing I’ve done (3500 runs; plus another 210 runs on a separate clinical test battery),** it seems that model family + ctx size dominate hallucination risk.**

  Eg: Qwen3-4B Hivemind ablation shows strong hallucination suppression (1.4% → 0.2% over 1000 runs) when context grounded. But it comes with a measured tradeoff: contradiction handling suffers under the constraints (detection metrics 2.00 → 0.00). When I ported the same routing policy to base Qwen 3-4B 2507 instruct, the gains flipped. No improvement, and format retries spiked to 24.9%. Still validating these numbers across conditions; still trying to figure out the why.

  For context, I tested:

  Reversal: Does the model change its mind when you flip the facts around? Or does it just stick with what it said the first time?

  Theory of Mind (ToM): Can it keep straight who knows what? Like, “Alice doesn’t know this fact, but Bob does” - does it collapse those into one blended answer or keep them separate?

  Evidence: Does it tag claims correctly (verified from the docs, supported by inference, just asserted)? And does it avoid upgrading vague stuff into false confidence?

  Retraction: When you give it new information that invalidates an earlier answer, does it actually incorporate that or just keep repeating the old thing?

  Contradiction: When sources disagree, does it notice? Can it pick which source to trust? And does it admit uncertainty instead of just picking one and running with it?

  Negative Control: When there’s not enough information to answer, does it actually refuse instead of making shit up?

  Using this as the source doc -

  tinyurl.com/GuardianMuskArticle

  FWIW, all the raw data, scores, and reports are here: codeberg.org/BobbyLLM/llama-conductor/…/prepub

  The Arxiv paper confirms what I’m seeing in the weeds: grounding and fabrication resistance are decoupled. You can be good at finding facts and still make shit up about facts that don’t exist. And Jesus, the gap between best and worst model at 32K is 70 percentage points? Temperature tuning? Maybe 2-3 pp gain. I know which lever I would be pulling (hint: pick a good LLM!),

  For clinical deployment under human review (which is my interest), I can make the case that trading contradiction flexibility for refusal safety is ok - it assumes the human in the middle reads the output and catches the edge cases.

  But if you’re expecting one policy to work across all models, automagically, you’re gonna have a bad time.

  TL;DR: context length is the primary degradation driver; my gut feeling based on the raw data here is that the useful window for local 4B is tighter ~16K. Above that hallucination starts to creep in, grounding or not.

  PS: I think (no evidence yet) that ablit and non abilt might need different grounding policies for different classes of questions. That’s interesting too - it might mean we can route between deterministic grounding and not differently, depending on ablation, to get the absolute best hallucination suppression. I need to think more on it.

  source

  • Womble@piefed.world ⁨2⁩ ⁨weeks⁩ ago

    I wouldnt read too much into the lower scores, they include some absolutely tiny models. The one 70% lower than the top score at 24% correct is a 1B model from 2024. Honestly that it can do any information retrival from a 32k context is impressive.

    source
  • how_we_burned@lemmy.zip ⁨2⁩ ⁨weeks⁩ ago

    I understood a few of those words.

    Basically you’ve validated the study that LLMs make shit up.

    source

    • SuspciousCarrot78@lemmy.world ⁨2⁩ ⁨weeks⁩ ago

      Well…no. But also yes :)

      Mostly, what I’ve shown is if you hold a gun to its head (“argue from ONLY these facts or I shoot”) certain classes of LLMs (like the Qwen 3 series I tested; I’m going to try IBM’s Granite next) are actually pretty good at NOT hallucinating, so long as 1) you keep the context small (probably 16K or less? Someone please buy me a better pc) and 2) you have strict guard-rails. And - as a bonus - I think (no evidence; gut feel) it has to do with how well the model does on strict tool calling benchmarks. Further, I think abliteration makes that even better.

      If that’s true (big IF), the we can reasonably quickly figure out (by proxy) which LLM’s are going to be less bullshitty when properly shackled.

      I’ll keep squeezing the stone until blood pours out. Stubbornness opens a lot of doors.

      source

      • -> View More Comments
• unpossum@sh.itjust.works ⁨2⁩ ⁨weeks⁩ ago

  GLM 4.5 is from August. Isn’t the real tl;dr that a seven month old open model, which was behind proprietary models at the time, did better than most humans would?

  source

  • MHard@lemmy.world ⁨2⁩ ⁨weeks⁩ ago

    The task described in this article is asking questions about a document that was provided to the llm in the context.

    I would hope that if you give a human a text and ask them to cite facts from it they would do better than 99% correct.

    Also, when the tokens exceeded 200k, the llm error rate was higher than 10%

    source

    • unpossum@sh.itjust.works ⁨2⁩ ⁨weeks⁩ ago

      I would hope that if you give a human a text and ask them to cite facts from it they would do better than 99% correct.

      That’s literally what school exams are about, isn’t it?

      Token window is a problem for all llms though, that’s not easily solved, but it can be worked around to a certain extent.

      source
• CubitOom@infosec.pub ⁨2⁩ ⁨weeks⁩ ago

  I’m not good at math, so someone please help me.

  If a model hallucinates 1% of the time for every question in a chat window that has 100 prompts in it, what is the chance of receiving a hallucination at some point in the chat?

  source

  • hersh@literature.cafe ⁨2⁩ ⁨weeks⁩ ago

    If I understand you correctly: 63.4% odds of having at least one hallucination.

    The simple way to calculate the odds of getting at least one error is to calculate the odds of having ZERO, and then inverting that.

    If the odds of a single instance being an error is 1%, that means you have a 99% chance of having no errors. If you repeat that 100 times, then it’s 99% of 99% of 99%…etc. In other words, 0.99^100 = 0.366. That’s the odds of getting zero errors 100 times in a row. The inverse of that is 0.634, or 63.4%.

    This is the same way to calculate the odds of N coin flips all coming up heads. It’s going to be 0.5^N. So the odds of getting 10 heads in a row is 0.5^10 = ~0.0977%, or 1:1024.

    source

    • CubitOom@infosec.pub ⁨2⁩ ⁨weeks⁩ ago

      Thanks, I also wonder how context collapse affects the fabrication rate.

      source
  • Telorand@reddthat.com ⁨2⁩ ⁨weeks⁩ ago

    One in 100. However, that is simple a measure of probability, so do not expect that to always be true for every 100 prompts.

    For example, if you rolled a 100-sided die 100 times, it’s possible to get a one every time. In practice, it would likely be a mix. You might have a session where you get no wrong answers and times when you get several.

    The problem is that ignorant people trust these models implicitly, because they sound convincing and authoritative, and many people are not equipped to be able to vet the information being generated (also notice I didn’t say “retrieved”).

    source
• rekabis@lemmy.ca ⁨2⁩ ⁨weeks⁩ ago

  How much do large language models actually hallucinate when answering questions grounded in provided documents?

  Okay, this is looking promising. In terms of the most important qualifications being plainly stated in the opening line.

  Because the amount of hallucinations/inaccuracies “in the wild” - depending on the model being tested - runs about 60-80%. But then again, this would be average use on generalized data sets, not questions on specific documentation. So of course the “in the wild” questions will see a higher rate.

  This also helps users, as it shows that hallucinations/inaccuracies can be reduced by as much as ⅔ by simply limiting LLMs to specific documentation that the user is certain contains the desired information.

  Very interesting!

  source

  • HubertManne@piefed.social ⁨2⁩ ⁨weeks⁩ ago

    I have been saying this for awhile. I am sorta hoping we see open source llms that are trained on a curated list of literature. its funny that these came out and it seemed like the makers did not take the long known garbage in - garbage out into account.

    source
  • SuspciousCarrot78@lemmy.world ⁨2⁩ ⁨weeks⁩ ago

    As I mentioned elsewhere (below) I am currently conducting similar testing across 4 different 4B models (Qwen 3-4B Hivemind, Qwen 3-4B 2507 Instruct, Phi-4-mini, Granite-4-h-micro), using both grounded and ungrounded states. I’m aiming for 10,000 runs (currently at 3500). Not to count chickens before they hatch but at ctx of 8192, it’s very much trending closer to 98-99.9% hallucination reduction for the class of questions asked (see prior post).

    If that holds (testing should be completed by tomorrow) that’s a useful thing to know. Likewise, if it doesn’t hold, also useful to know.

    I actually have an idea of how to make this even better / more useful. Again, chicken have not hatched. I will share what I discover here if there is interest / submit for peer review.

    source
• Son_of_Macha@lemmy.cafe ⁨2⁩ ⁨weeks⁩ ago

  We need to stop calling it hallucinations and say what it is, ERRORS

  source

  • cmhe@lemmy.world ⁨2⁩ ⁨weeks⁩ ago

    Hallucinations of LLMs are just one class of errors, and the most dangerous one.

    Other stuff like garbeled or repeating output are other errors.

    source
• HubertManne@piefed.social ⁨2⁩ ⁨weeks⁩ ago

  This is why I would encourage people to use llms for something not important. like video games or interests. You likely will have enough knowledge around the things to catch the “hallucinations” and hopefully that will give you perspective on their use for more important things.

  source

  • Son_of_Macha@lemmy.cafe ⁨2⁩ ⁨weeks⁩ ago

    Or not at all

    source

    • HubertManne@piefed.social ⁨2⁩ ⁨weeks⁩ ago

      see if they don’t at all then they can fall victim to thinking they are better than they are. By using it a bit in something unimportant which you are knowledgable enough about it allows you to see the flaws and it does not take that much time to see them.

      source

      • -> View More Comments
• FrankLaskey@lemmy.ml ⁨2⁩ ⁨weeks⁩ ago

  My biggest takeaway here is that choosing the context length and (to a lesser extent) the temperature carefully is important for reducing hallucinations. I expected model families to vary widely between themselves but not for context length to have such a massive impact tbh.

  It seems from this like reducing context length in applications where it isn’t essential for the model to hold very large amounts of context simultaneously would be best practice no?

  source
• fallaciousBasis@lemmy.world ⁨2⁩ ⁨weeks⁩ ago

  People call it hallucinating but it seems pretty much identical to rationalization.

  source