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%.
I understood a few of those words.
Basically you’ve validated the study that LLMs make shit up, right?
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.
Are all outputs hallucinations? It’s just some happen to be correct and some aren’t. It doesn’t know and can’t tell unless it’s specifically told (hence the guard rails).
But if I’ve gotta build so many hand rails (instructions) then is it really “AI”?
Point 1 - no. LLM outputs are not always hallucinations (generally speaking - some are worse than others) but where they might veer off into fantasy, I’ve reinforced with programming. Think of it like giving your 8yr old a calculator instead of expecting them to work out 7532x565 in their head. And a dictionary. And encyclopedia. And Cliff’s notes. And watch. And compass. And a … you get the idea.
The role of the footer is to show you which tool it used (its own internal priors, what you taught it, calculator etc) and what ratio the answer is based on those. Those are router assigned. That’s just one part of it though.
Point 2 is a mis-read. These aren’t instructions or system prompts telling the model “don’t make things up” - that works about as well as telling a fat kid not to eat cake.
Instead, what happens is the deterministic elements fire first. The model gets the answer, which the model then builds context on. It funnels it in the right direction and the llm tends to stay in that lane. That’s not guardrails on AI, that’s just not using AI where AI is the wrong tool. Whether that’s “real AI” is a philosophy question - what I do know and can prove is that it leads to far fewer wrong answers.
EDIT: I got my threads mixed. Still same point but for context, see - https://lemmy.world/post/44805995
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!
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
Like a certain famous space Lothario, I too do not believe in no win scenarios.
Obligatory reference since you mention AI and no win scenarios: https://www.msn.com/en-in/news/India/claude-discovers-the-kobayashi-maru-test-what-is-the-benchmark-safety-test-the-ai-chatbot-outsmarted/ar-AA1YqyEY