Abstract:
Hallucination has been widely recognized to be a significant drawback for large language models (LLMs). There have been many works that attempt to reduce the extent of hallucination. These efforts have mostly been empirical so far, which cannot answer the fundamental question whether it can be completely eliminated. In this paper, we formalize the problem and show that it is impossible to eliminate hallucination in LLMs. Specifically, we define a formal world where hallucina- tion is defined as inconsistencies between a computable LLM and a computable ground truth function. By employing results from learning theory, we show that LLMs cannot learn all of the computable functions and will therefore always hal- lucinate. Since the formal world is a part of the real world which is much more complicated, hallucinations are also inevitable for real world LLMs. Furthermore, for real world LLMs constrained by provable time complexity, we describe the hallucination-prone tasks and empirically validate our claims. Finally, using the formal world framework, we discuss the possible mechanisms and efficacies of existing hallucination mitigators as well as the practical implications on the safe deployment of LLMs.
I extremely doubt that hallucination is a limitation in final output. It may be an inevitable part of the process, but it’s almost definitely a surmountable problem.
Just off the top of my head I can imagine using two separate LLMs for a final output, the first one generates an initial output, and the second one verifies whether what it says is accurate. The chance of two totally independent LLMs having the same hallucination is probably very low. And you can add as many additional separate LLMs for re-verification as you like. The chance of a hallucination making it through multiple LLM verifications probably gets close to zero.
While this would greatly multiply the resources required, it’s just a simple example showing that hallucinations are not inevitable in final output
That’s not how LLMs work.
Super short version is that LLMs probabilistically determine the next word most likely to occur in a sequence. They do this using Statistical Models (like what your cell phone’s auto complete uses); Transformers (rating the importance of preceding words, so the model can “focus” on the most important words); and Relatedness (a measure of how closely linked different words/phrases are to reach other in meaning).
With increasingly large models, LLMs can build a more accurate representation of Relatedness across a wider range of topics. With enough examples, LLMs can infinitely generate content that is closely Related to a query.
So a small LLM can make sentences that follow writing conventions but are nonsense. A larger LLM can write intelligibly about topics that are frequently included in the training materials. Huge LLMs can do increasingly nuanced things like “explain” jokes.
LLMs are not capable of evaluating truth or facts. It’s not part of the algorithm. And it doesn’t matter how big they get. At best, with enough examples to build a stronger Relatedness dataset, they are more likely to “stay on topic” and return results that are actually similar to what is being asked.
No, I’ve used LLMs to do exactly this, and it works. You prompt it with a statement and ask “is this true, yes or no?” It will reply with a yes or no, and it’s almost always correct. Do this verification through multiple different LLMs and it would eliminate close to 100% of hallucinations.
EDIT
I just tested it multiple times in chatgpt4, and it got every true/false answer correct.
This can only work in contexts where the LLM already has a strong Relatedness database for the topic. LLMs are incapable of assessing the accuracy of any information they weren’t trained on with sufficient examples to build that database.
This paper, if I’m understanding it correctly, is saying the same thing: LLMs will always hallucinate and are incapable of identifying hallucinations in certain contexts.
The problem we face is that it’s hard to identify hallucinations in the exact context when LLMs are most likely to make them: in content that’s not widely known and understood with many examples in the training data.
Or maybe I’m off base and I need to read the full study.
There are far more important facets to truthfulness and semantics than yes/no questions. If this is the only way you evaluate LLM’s, you will quickly fall for confirmation bias.
two totally independent LLMs
How do you propose to get these independent LLMs? If both are trained using similar objectives e.g., masked token prediction, then they won’t be independent.
Also, assuming independent LLMs could be obtained, how do you propose to compute this hallucination probability? Without knowing this probability, you can’t know how many verification LLMs are sufficient for your application, can you?
There are already existing multiple different LLMs that are essentially completely different. In fact this is one of the major problems with LLMs, because when you add even a small amount of change into an LLM it turns out to radically alter the output it returns for huge amounts of seemingly unrelated topics.
For your other point, I never said bouncing their answers back and forth for verification was trivial, but it’s definitely doable.
Can you provide the source of a few of these completely different LLMs?
add even a small amount of change into an LLM […] radically alter the output
You mean perturbing the parameters of the LLM? That’s hardly surprising IMO. And I’m not sure it’s convincing enough to show independence, unless you have a source for this?
even a small amount of change into an LLM it turns out to radically alter the output it returns for huge amounts of seemingly unrelated topics.
Do you mean that small changes radically change the phrasing of answers, but that it has largely the same “knowledge” of the world? Or do you mean that small changes also radically alter what a llm thinks is true or not? If you think the former is true, then these models should still be the same in regards to what they think is true or not, and if you don’t then you think that llms perception of the world is basically arbitrary and in that case we shouldn’t trust them to tell us what’s true at all.
