this post was submitted on 11 Mar 2026
110 points (97.4% liked)

Technology

82516 readers
5335 users here now

This is a most excellent place for technology news and articles.


Our Rules


  1. Follow the lemmy.world rules.
  2. Only tech related news or articles.
  3. Be excellent to each other!
  4. Mod approved content bots can post up to 10 articles per day.
  5. Threads asking for personal tech support may be deleted.
  6. Politics threads may be removed.
  7. No memes allowed as posts, OK to post as comments.
  8. Only approved bots from the list below, this includes using AI responses and summaries. To ask if your bot can be added please contact a mod.
  9. Check for duplicates before posting, duplicates may be removed
  10. Accounts 7 days and younger will have their posts automatically removed.

Approved Bots


founded 2 years ago
MODERATORS
 

Evaluating 35 open-weight models across three context lengths (32K, 128K, 200K), four temperatures, and three hardware platforms—consuming 172 billion tokens across more than 4,000 runs—we find that the answer is “substantially, and unavoidably.” Even under optimal conditions—best model, best temperature, temperature chosen specifically to minimize fabrication—the floor is non-zero and rises steeply with context length. At 32K, the best model (GLM 4.5) fabricates 1.19% of answers, top-tier models fabricate 5–7%, and the median model fabricates roughly 25%.

top 37 comments
sorted by: hot top controversial new old
[–] SuspciousCarrot78@lemmy.world 3 points 3 hours ago* (last edited 3 hours ago) (2 children)

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. Yes, a real "science discovers water makes things wet; news at 11" moment.

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 -

https://tinyurl.com/GuardianMuskArticle

FWIW, all the raw data, scores, and reports are here: https://codeberg.org/BobbyLLM/llama-conductor/src/branch/main/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: once you control for model family, I think context length is going to turn out the be the main 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. It would be neat if it was a simple 4x relationship (4b-->16K; 8b-->32K) but things tend not to work out that nicely IRL.

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.

[–] Womble@piefed.world 2 points 2 hours 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.

[–] how_we_burned@lemmy.zip 1 points 2 hours ago* (last edited 2 hours ago) (1 children)

I understood a few of those words.

Basically you've validated the study that LLMs make shit up, right?

[–] SuspciousCarrot78@lemmy.world 2 points 2 hours ago* (last edited 1 hour ago) (1 children)

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. Let me find out.

If any of that's true (big IF), then we can reasonably quickly figure out (by proxy) which LLM's are going to be less bullshitty when properly shackled, in every day use. For reference, Qwen 3 and IBM Granite (both of which have abliterated version IIRC - that is, safety refusals removed) are known to score highly on tool calling. 4 swallows don't make spring but if someone with better gear wants to follow that path, then at least I can give some prelim data from the potato frontier.

I'll keep squeezing the stone until blood pours out. Stubbornness opens a lot of doors. I refuse to be told this is an intractable problem; at least until I try to solve it myself.

[–] andallthat@lemmy.world 1 points 58 minutes ago (1 children)

is "potato frontier" an auto-correct fail for Pareto or a real term? Because if it's not a real term, I'm 100% going to make it one!

[–] SuspciousCarrot78@lemmy.world 1 points 5 minutes ago* (last edited 3 minutes ago)

No, it's real. I'm running on a Quadro P1000 with 4GB vram (or a Tesla P4 with 8GB). My entire raison d'être is making potato tier computing a thing.

https://openwebui.com/posts/vodka_when_life_gives_you_a_potato_pc_squeeze_7194c33b

That and because fuck Chatgpt

I refuse to accept no win scenarios.

[–] fallaciousBasis@lemmy.world 1 points 5 hours ago

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

[–] RandAlThor@lemmy.ca 14 points 19 hours ago (3 children)

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

[–] Zink@programming.dev 5 points 14 hours 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!

[–] bad1080@piefed.social 13 points 18 hours ago (1 children)
[–] snooggums@piefed.world 8 points 18 hours ago (2 children)

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.

[–] Scipitie@lemmy.dbzer0.com 18 points 18 hours ago (1 children)

Accepting concepts like "right" and "wrong" gives those tools way too much credit, basically following the AI narrative of the corporations behind them. They can only be used about the output but not the tool itself.

To be precise:

LLMs can't be right or wrong because the way they work has no link to any reality - it's stochastics, not evaluation. I also don't like the term halluzination for the same reason. It's simply a too high temperature setting jumping into a closeby but unrelated vector set.

Why this is an important distinction: Arguing that an LLM is wrong is arguing on the ground of ChatGPT and the likes: It's then a "oh but wen make them better!" And their marketing departments overjoy.

To take your calculator analogy: like these tools do have floating point errors which are inherent to those tools wrong outputs are a dore part of LLMs.

We can minimize that but then they automatically use part of their function. This limitation is way stronger on LLMs than limiting a calculator to 16 digits after the comma though...

[–] CubitOom@infosec.pub 5 points 17 hours ago* (last edited 17 hours ago) (3 children)

What word would you propose to use instead?

Fabrication?

[–] eceforge@discuss.tchncs.de 2 points 6 hours ago

No comment on the rest of the thread but I always though "confabulation" was a more accurate word than hallucination for what LLMs tend to do.

The "signs and symptoms" part of the article really seems oddly familiar when compared to interacting with an LLM sometimes haha.

[–] Scipitie@lemmy.dbzer0.com 7 points 17 hours ago (1 children)

That's my problem: any single word humanizes the tool in my opinion. Iperhaps something like "stochastic debris" comes close but there's no chance to counter the common force of pop culture, Corp speak a and humanities talent to see humanoid behavior everywhere but each other. :(

[–] Telorand@reddthat.com 4 points 17 hours ago (1 children)

We do enjoy pareidolia, don't we?

[–] deranger@sh.itjust.works 2 points 15 hours ago (1 children)

Paredolia just means seeing patterns that aren’t there, it’s not implicitly human. If you see a dog in the clouds, that’s paredolia.

[–] Telorand@reddthat.com 1 points 1 hour ago (1 children)

Great, when did I say otherwise? Pareidolia is a thing humans do, because we like patterns. Finding patterns is something that has benefited our species, but it is sometimes so strong that we see faces in electrical outlets or the shape of a car's front profile (for example).

[–] deranger@sh.itjust.works 1 points 53 minutes ago* (last edited 53 minutes ago)

I mean, it doesn’t really follow given the context. Nobody is talking about the visual sense, they’re talking about humanizing AI through using certain words, which isn’t paredolia.

[–] leftzero@lemmy.dbzer0.com 2 points 13 hours ago

Scam. We're being sold an autocomplete tool as a search engine.

Or fraud, since some of the same companies destroyed the functionality of their search engines in order to make the autocomplete look better in comparison.

[–] bad1080@piefed.social 4 points 18 hours ago

if you have a lobby you get special names, look at the pharma industry who coined the term "discontinuation syndrome" for a simple "withdrawal"

[–] ji59@hilariouschaos.com 2 points 18 hours ago* (last edited 18 hours ago)

Because guessing correct answer is more successful than saying nothing.

[–] CubitOom@infosec.pub 6 points 17 hours ago (2 children)

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?

[–] hersh@literature.cafe 14 points 16 hours ago* (last edited 16 hours ago) (1 children)

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.

Edit: This is assuming independence of all 100 prompts, which is not generally true in a single chat window, where each prompt follows the last and retains both the previous prompts and answers in its context. As the paper explains, error rate tends to increase with context length. You should generally start a new chat rather than continue in an existing one if the previous context is not highly relevant.

[–] CubitOom@infosec.pub 4 points 16 hours ago

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

[–] Telorand@reddthat.com 0 points 17 hours 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").

[–] FauxLiving@lemmy.world 2 points 18 hours ago (2 children)

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.

[–] snooggums@piefed.world 8 points 18 hours ago* (last edited 18 hours ago) (1 children)

Calculators are correct 100% of the time.

[–] FauxLiving@lemmy.world 4 points 17 hours ago (1 children)

Calculators are not people, Mr. <1.19%.

[–] snooggums@piefed.world 7 points 17 hours ago* (last edited 16 hours ago) (2 children)

That's right! We should be comparing computers to computers. Well, hardware computers, not people computers.

[–] FauxLiving@lemmy.world 4 points 16 hours ago

Calculators are not computers, computers contain calculator-like elements but a calculator is no more a computer than a passenger jet is a coffee shop by virtue of having a coffee pot onboard.

Calculators cannot fabricate answers, but nor are they 100% correct due to things like bitflips and square root approximations. They also cannot write text, so the comparison would make even less sense.

LLMs and Humans can fabricate answers in written text so comparing the fabrication rate in written text of an LLM to a human (both entities which generate their answers with neural networks) makes more sense than to compare either to a calculator which neither uses a neural network or produces text.

So 'we' should compare like things and not choose items based on superficial similarities.

[–] ji59@hilariouschaos.com 0 points 16 hours ago

What do you even mean? Calculators and LLMs are solving different problems. And there are a lot of calculators and a lot of LLMs. Also, calculator accuracy could be approaching 0% because they all have limited precision and there are infinite numbers. Some of the calculators even can't correctly answer 0.1+0.2, while most LLMs can do that.

[–] Iconoclast@feddit.uk 3 points 16 hours ago* (last edited 16 hours ago) (1 children)

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

[–] FauxLiving@lemmy.world 2 points 13 hours ago

Truly the most humble person of all time.

[–] unpossum@sh.itjust.works 0 points 18 hours ago (1 children)

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?

[–] MHard@lemmy.world 9 points 14 hours ago (1 children)

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%

[–] unpossum@sh.itjust.works 2 points 7 hours 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.