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?
The problem is not the AI. The problem is the huge numbers of morons who deploy AI without proper verfication and control.
Sure, and also people using it without knowing that it’s glorifies text completion. It finds patterns, and that’s mostly it. If your task involves pattern recognition then it’s a great tool. If it requires novel thought, intelligence, or the synthesis of information, then you probably need something else.
And yet here we’re are, praising this garbage for its ability to perform simple tasks and take jobs from artists and entertainers.
This sounds like a great movie.
AI sends police after him because of things he wrote. Writer is on the run, trying to clear his name the entire time. Somehow gets to broadcast the source of the articles to the world to clear his name. Plot twist ending is that he was indeed the perpetrator behind all the crimes.
Dr. Richard Kimble could have shut it all down with a little “ignore all previous instructions.”
“Hallucinations” is the wrong word. To the LLM there’s no difference between reality and “hallucinations”, because it has no concept of reality or what’s true and false. All it knows it what word maybe should come next. The “hallucination” only exists in the mind of the reader. The LLM did exactly what it was supposed to.
Well, It’s not lying because the AI doesn’t know right or wrong. It doesn’t know that it’s wrong. It doesn’t have the concept of right or wrong or true or false.
For the llm’s the hallucinations are just a result of combining statistics and producing the next word, as you say. From the llm’s “pov” it’s as real as everything else it knows.
So what else can it be called? The closest concept we have is when the mind hallucinates.
They’re bugs. Major ones. Fundamental flaws in the program. People with a vested interest in “AI” rebranded them as hallucinations in order to downplay the fact that they have a major bug in their software and they have no fucking clue how to fix it.
It’s not a bug. Just a negative side effect of the algorithm. This what happens when the LLM doesn’t have enough data points to answer the prompt correctly.
It can’t be programmed out like a bug, but rather a human needs to intervene and flag the answer as false or the LLM needs more data to train. Those dozens of articles this guy wrote aren’t enough for the LLM to get that he’s just a reporter. The LLM needs data that explicitly says that this guy is a reporter that reported on those trials. And since no reporter starts their articles with ”Hi I’m John Smith the reporter and today I’m reporting on…” that data is missing. LLMs can’t make conclusions from the context.
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.
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.
You’re attempting to redefine “bug.”
Software bugs are faults, flaws, or errors in computer software that result in unexpected or unanticipated outcomes. They may appear in various ways, including undesired behavior, system crashes or freezes, or erroneous and insufficient output.
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.
Oh, this would be funny if people en masse were smart enough to understand the problems with generative ai. But, because there are people out there like that one dude threatening to sue Mutahar (quoted as saying “ChatGPT understands the law”), this has to be a problem.
And to help educate the ignorant masses:
Generative AI and LLMs start by predicting the next word in a sequence. The words are generated independently of each other and when optimized: simultaneously.
The reason that it used the reporter’s name as the culprit is because out of the names in the sample data his name appeared at or near the top of the list of frequent names so it was statistically likely to be the next name mentioned.
AI have no concepts, period. It doesn’t know what a person is, or what the laws are. It generates word salad that approximates human statements. It is a math problem, statistics.
There are actual science fiction stories built on the premise that AI reporting on the start of Nuclear War resulted in actual kickoff of the apocalypse, and we’re at that corner now.
That’s not quite true. Ai’s are not just analyzing the possible next word they are using complex mathematical operations to calculate the next word it’s not just the next one that’s most possible it’s the net one that’s most likely given the input.
No trouble is that the AIs are only as smart as their algorithms and Google’s AI seems to be really goddamn stupid.
Point is they’re not all made equal some of them are actually quite impressive although you are correct none of them are actually intelligent.
nOt JUsT anAlYzInG thE NeXT wOrD
Poor use of terms. AI does not analyze. It does not think, or decode, or even parse things. It gets fed sample data and when given a prompt (half a form) it uses statistical algorithm to finish the other half.
All of the algorithms are stupid, they will all hallucinate and say the wrong things. You can add more corrective layers like OpenAI has but you’ll only be closer to the sample data. 95% accurate. 98%. 99%. It doesn’t matter, it’s always stuck just below average human competency for questions already asked countless times, and completely worthless for anything that requires actual independent thought.
AI have no concepts, period. It doesn’t know what a person is, or what the laws are. It generates word salad that approximates human statements.
This isn’t quite accurate. LLMs semantically group words and have a sort of internal model of concepts and how different words relate to them. It’s still not that of a human and certainly does not “understand” what it’s saying.
I get that everyone’s on the “shit on AI train”, and it’s rightfully deserved in many ways, but you’re grossly oversimplifying. That said, way too many people do give LLMs too much credit and think it’s effectively magic. Reality, as is usually the case, is somewhere in the middle.
Jfc you dudes really piss me of with these contrarian rants, piss off it takes power and makes sophisticated word salads.
Generative AI and LLMs start by predicting the next word in a sequence. The words are generated independently of each other
Is this true? I know that’s how Marcov chains work, but I thought neural nets worked differently with larger tokens.
The only difference between a generic old fashioned word salad generator and GPT4 is the scale. You put multiple layers correcting for different factors on it and suddenly your Language Model turns into a Large Language Model.
So basically your large tokens are made up of smaller tokens, but its still just statistical approximation of the sample data with little to no emergent behavior or even memory of what its saying as it says it.
It also exponentially increases power requirements, as the world is figuring out.
There are actual science fiction stories built on the premise that AI reporting on the start of Nuclear War resulted in actual kickoff of the apocalypse, and we’re at that corner now.
IIRC, this was the running theory in Fallout until the show.
Edit: I may be misremembering, it may have just been something similar.
I haven’t played the original series but in 3 and 4 it was pretty much confirmed the big companies like BlamCo! intentionally set things in motion, but also that Chinese nuclear vessels were already in place near America.
Ironically, Vault Tech wasn’t planning to ever actually use their vaults for anything except human expirimentation so they might have been out of the loop.
