When German journalist Martin Bernklautyped his name and location into Microsoft’s Copilot to see how his articles would be picked up by the chatbot, the answers horrified him. Copilot’s results asserted that Bernklau was an escapee from a psychiatric institution, a convicted child abuser, and a conman preying on widowers. For years, Bernklau had served as a courts reporter and the AI chatbot had falsely blamed him for the crimes whose trials he had covered.
The accusations against Bernklau weren’t true, of course, and are examples of generative AI’s “hallucinations.” These are inaccurate or nonsensical responses to a prompt provided by the user, and they’re alarmingly common. Anyone attempting to use AI should always proceed with great caution, because information from such systems needs validation and verification by humans before it can be trusted.
But why did Copilot hallucinate these terrible and false accusations?
It’s an inherent negative property of the way they work. It’s a problem, but not a bug any more than the result of a car hitting a tree at high speed is a bug.
Calling it a bug indicates that it’s something unexpected that can be fixed, and as far as we know it can’t be fixed, and is expected behavior. Same as the car analogy.
The only thing we can do is raise awareness and mitigate.
You’re attempting to redefine “bug.”
From a software testing point of view, a correctly coded realization of an erroneous algorithm is a defect (a bug). It fails validation (a test for fitness for use) rather than verification (a test that the code correctly implements the erroneous algorithm).
This kind of issue arises not only with LLMs, but with any software that includes some kind of model within it. The provably correct realization of a crap model is still crap.
It actually can be fixed. There is an accuracy to answers. Like how confident the statistical model is on the answer. That’s why some questions get consistent answers while others don’t.
The fix is not that hard, it’s a matter of reputation on having the chatbot answer “I don’t know” when the confidence on an answer isn’t high enough. It’s pretty similar on what the chatbot does when you ask them to make you a bomb, just highjacks the answer calculated by the model and says a predefined answer instead.
But it makes the AI look bad. So most public available models just answer anything even if they are not confident about it. Also your reaction to the incorrect answer is used to train the model better so it’s not even efficient for they to stop the hallucinations on their product. But it can be done.
Models used by companies usually have a higher confidence threshold and answer “I don’t know” if they don’t have enough statistical proof on a particular answer.
This has been tried, it’s helping but it’s not enough by itself. It’s one of the mitigation steps I was thinking of. And companies do work very hard to reduce hallucinations, just look at Microsoft’s newest thing.
From that article:
The Hidrogen from water thing is simply wrong. If that is supposed to mean that hallucinations are just part of a generative LLM technology that cannot be solved.
They are not inherent of the technology. They are a product of lack of control over the stadistical output. Prioritizing any answer before no answer.
As with any statistics you have a confidence on how true something is based on your data. It’s just a matter of putting the threshold higher or lower.
If you ask an easy question “What is the capital of France?” You wont ever get an hallucination. Because all models will have that answer provided with very high confidence. You just have to make so if that level of confidence is not reached it just default to a “I don’t know answer”. But, once again, this will make the chatbots seem very dumb as they will answer with lots of “I don’t know”.
The problem here is the amount of data and the efficiency of the model. In order to get an usable general purpose model with a confidence threshold high enough to not hallucinate, by todays efficiency with the models it would need to be an humongous model, too big and with too much training data even for big tech. So we can go that big, we can try to improve efficiency (which is being proven very hard for general models) or we do both. Time will tell, but I’m quite confident that we will reach a general use model without hallucinations sooner or later.
I think you misunderstand how LLM’s work, it doesn’t have a confidence, it’s not like it looks at it’s data and say “hmm, yes, most say Paris is the capital of France, so that’s the answer”. It “just” puts weight on the next token depending on it’s internal statistics, and then one of those tokens are picked, and the process start anew.
Teaching the model to say “I don’t know” helps a bit, and was lauded as “The Solution” a year or two ago but turns out it didn’t really help that much. Then you got Grounded approach, RAG, CoT, and so on, all with the goal to make the LLM more reliable. None of them solves the problem, because as the PhD said it’s inherent in how LLM’s work.
And no, local llm’s aren’t better, they’re actually much worse, and the big companies are throwing billions on trying to solve this. And no, it’s not because “that makes the llm look dumb” that they haven’t solved it.
Early on I was looking into making a business of providing local AI to businesses, especially RAG. But no model I tried - even with the documents being part of the context - came close to reliable enough. They all hallucinated too much. I still check this out now and then just out of own interest, and while it’s become a lot better it’s still a big issue. Which is why you see it on the news again and again.
This is the single biggest hurdle for the big companies to turn their AI’s from a curiosity and something assisting a human into a full fledged autonomous / knowledge system they can sell to customers, you bet your dangleberries they try everything they can to solve this.
And if you think you have the solution that every researcher and developer and machine learning engineer have missed, then please go prove it and collect some fat checks.
What do you think is “weight”?
Is, simplifying, the amounts of data that says “The capital of France is Paris” it doesn’t need to understand anything. It just has to stop the process if the statistics don’t not provide enough to continue with confidence. If the data is all over the place and you have several “The capital of France is Berlin/Madrid/Milan”, it’s measurable compared to all data saying it is Paris. Not need for any kind of “understanding” of the meaning of the individual words, just measuring confidence on what next word should be.
Back a couple of years when we played with small neural networks playing mario and you could see the internal process in real time, as there where not that many layers. It was evident how the process and the levels of confidence changed depending on how deep the training was. Here it is just orders of magnitude above. But nothing imposible to overcome as some people pretend to sell.
Alternative ways of measure confidence is just run the same question several times and check if answers are equivalent.
PhD is PhD in scaremongering about technology, so it’s not an authority on anything here.
IDK what did you do, but slm don’t really hallucinate that much, if at all. Specially if they are trained with good datasets.
As I said the solution is not in my hand, as it involves improving the efficiency or the amount of data. Efficiency has issues as current techniques seems to be unable to improve efficiency over a certain level. And amount of data is, obviously, costly.
You can call that confidence if you want, but it got very little to do with how “sure” the model is.
Actually, it would be "The confidence of token Th is 0.95, the confidence of S is 0.32, the confidence of … " and so on for each possible token, many llm’s have around 16k-32k token vocabulary. Most will be at or near 0. So you pick Th, and then token “e” will probably be very high next, then a space token, then… Anyway, the confidence of the word “Paris” won’t be until far into the generation.
Now there is some overseeing logic in a way, if you ask what the capitol of a non existent country is it’ll say there’s no such country, but is that because it understands it doesn’t know, or the training data has enough examples of such that it has the statistical data for writing out such an answer?
I assume by SLM you mean smaller LLM’s like for example mistral 7b and llama3.1 8b? Well those were the kind of models I did try for local RAG.
Well, it was before llama3, but I remember trying mistral, mixtral, llama2 70b, command-r, phi, vicuna, yi, and a few others. They all made mistakes.
I especially remember one case where a product manual had this text : “If the same or a newer version of <product> is already installed on the computer, then the <product> installation will be aborted, and the currently installed version will be maintained” and the question was “What happens if an older version of <product> is already installed?” and every local model answered that then that version will be kept and the installation will be aborted.
When trying with OpenAI’s latest model at that time, I think 4, it got it right. In general, about 1 in ~5-7 answers to RAG backed questions were wrong, depending on the model and type of question. I could usually reword the question to get the correct answer, but to do that you kinda already have to know the answer is wrong. Which defeats the whole point of it.
More or less that. There’s a point during the path that the input is taking on the language model were the induced randomness can significantly affect the output or not. If all the weights are pointing to the same end node, because the “confidence” is high, the no matter the random seed, the output will be the same. When the seed greatly affect the final result is because the weights don’t point with that confidence to an unique end node, so the small randomness introduced at the beginning (the seed to say so) greatly change the result. It is here were you are most likely to get an hallucination.
To put again in terms of the much more easier to view earlier neural networks. When you didn’t trail the model enough mario just made random movements without doing attempts to complete the level. Because the weights of the neurons could not reliably take the input and transform into an useful output. It os something that could be solved in smaller models. For larger models gets incredibly complicated because the massive amount of data. The complexity of the data. And the complexity of a proper training. But it’s not something imposible or that could not get rid of. The same you can get Mario to finally complete all levels every time without issues, you can get a non hallucinanting chat bot, it just takes more technology improvements.
I suppose it could be said that the nature of language is chaotic like weather and not deterministic like a Mario level, and thus it would be actually “impossible” to get large results, like it’s impossible to get precise weather a month in advance. But I’m not sure there would be enough evidence to support that, as hallucinations are not just across the board, they just tend to happen on matters that had little training data. Matters with plenty of training data do not hallucinate even in today models.
I searched slm online and found out that small models you said. I wasn’t refering to those. Those are just small large language models IMO if that makes any sense. A proper slm should also have a small purpose, cannot be general chat. I mostly refer to the current chatbots that point you to predefined answers, or summarizing ones. Nothing that could really elaborate a wrote answer word by word.
Currently and to my knowledge. There isn’t any general language model that can just write up answers and that is good enough to not hallucinate. But certainly we are getting closer each year.
Edit: I’ve been looking for an example, here https://www.tax.service.gov.uk/ask-hmrc/chat/self-assessment These kind of chatbots, they know when their answer is not precise and default to a polite “ask again” answer instead of just tell you the first “hallucination” that came to them. They are powered by similar AI technology but it’s not a general use and cannot write word by word. But it “knows” when te answer is precise or not.
The example you shared is not an LLM. It’s a classic chatbot with pre-defined answers. It basically knows keyword to KB article. If no term is known, it will tell “I don’t know”. It will also suggest incorrect KB if picks one keyword, ignoring the rest of the context. It has no idea of the answer is correct by any means. At best somebody will periodically check a sample of questions that the user didn’t consider correct to evaluate the pairings, but it’s not AI, at least not a good one
This article is an example where statistical confidence doesn’t help. The model has lots of data so it likely has high confidence, but it didn’t have any understanding of the nature of the relation in the data.
I recently did an application where we indicated the confidence of the output of the model. For some scenarios, the high confidence output had even more mistakes than the low confidence output
OK, so describe how you control that output so that hallucinations don’t occur. Does the anti-hallucination training set exceed the size of the original LLM’s training set? How is it validated? If it’s validated by human feedback, then how much of that validation feedback is required, and how do you know that the feedback is not being used to subvert the model rather than to train it?