First, although GPT-4’s UBE score nears the 90th percentile when examining approximate conversions from February administrations of the Illinois Bar Exam, these estimates are heavily skewed towards repeat test-takers who failed the July administration and score significantly lower than the general test-taking population.
What I find delightful about this is that I already wasn’t impressed! Because, as the paper goes on to say
Moreover, although the UBE is a closed-book exam for humans, GPT-4’s huge training corpus largely distilled in its parameters means that it can effectively take the UBE “open-book”
And here I was thinking it not getting a perfect score on multiple-choice questions was already damning. But apparently it doesn’t even get a particularly good score!
[…W]hen examining only those who passed the exam (i.e. licensed or license-pending attorneys), GPT-4’s performance is estimated to drop to 48th percentile overall, and 15th percentile on essays.
officially Not The Worst™, so clearly AI is going to take over law and governments any day now
also. what the hell is going on in that other reply thread. just a parade of people incorrecting each other going “LLM’s don’t work like [bad analogy], they work like [even worse analogy]”. did we hit too many buzzwords?
Not the worst? 48th percentile is basically “average lawyer”. I don’t need a Supreme Court lawyer to argue my parking ticket. And if you train the LLM with specific case law and use RAG can get much better.
In a worst case scenario if my local lawyer can use AI to generate a letter and just quickly go through it to make sure it didn’t hallucinate, they can process more clients, offer faster service and cheaper prices. Maybe not a revolution but still a win.
good thing all of law is just answering multiple-choice tests
I don’t need a Supreme Court lawyer to argue my parking ticket.
because judges looooove reading AI garbage and will definitely be willing to work with someone who is just repeatedly stuffing legal-sounding keywords into google docs and mashing “generate”
And if you train the LLM with specific case law and use RAG can get much better.
“guys our keyword-stuffing techniques aren’t working, we need a system to stuff EVEN MORE KEYWORDS into the keyword reassembler”
In a worst case scenario if my local lawyer can use AI to generate a letter
oh i would love to read those court documents
and just quickly go through it to make sure it didn’t hallucinate
wow, negative time saved! okay so your lawyer has to read and parse several paragraphs of statistical word salad, scrap 80+% of it because it’s legalese-flavored gobbledygook, and then try to write around and reformat the remaining 20% into something that’s syntactically and legally coherent – you know, the thing their profession is literally on the line for. good idea
what promptfondlers continuously seem to fail to understand is that verification is the hard step. literally anyone on the planet can write a legal letter if they don’t care about its quality or the ramifications of sending it to a judge in their criminal defense trial. part of being a lawyer is being able to tell actual legal arguments from bullshit, and when you hire an attorney, that is the skill you are paying for. not how many paragraphs of bullshit they can spit out per minute
they can process more clients, offer faster service and cheaper prices. Maybe not a revolution but still a win.
“but the line is going up!! see?! sure we’re constantly losing cases and/or getting them thrown out because we’re spamming documents full of nonsense at the court clerk, but we’re doing it so quickly!!”
it’s funny how your first choice of insult is accusing me of not being deep enough into llm garbage. like, uh, yeah, why would i be
but also how dare you – i’ll have you know i only choose the most finely-tuned, artisinally-crafted models for my lawyering and/or furry erotic roleplaying needs
In a worst case scenario if my local lawyer can use AI to generate a letter and just quickly go through it to make sure it didn’t hallucinate, they can process more clients, offer faster service and cheaper prices.
It’s a good thing people are so good at vigilance tasks and don’t tend to fall onto just relying on the automation.
You understand that getting a list of sources and checking them is easier than finding them on your own, right?
Of course it’s even easier not checking them at all and submitting garbage, but one should have learned in 3rd grade not to submit copy-pastes from Wikipedia or any website.
This one is on human stupidity, not artifical intelligence.
ask chatgpt, which will output convincing blob of text, with references and sources that might or might be not real, relevant, or make sense, some of which you won’t be able to judge
then, ask a real lawyer about this, which means that they have to make sense of the situation on their own but also dig through machine generated drivel, which means that they need more time for that, and this means extra cost/wasted effort
You understand that getting a list of sources and checking them is easier than finding them on your own, right?
that’s one weirdass assumption. when you know what are you looking for, the opposite is true. few months back i’ve authored a review chapter in my (very narrow) field, and while “getting a list of sources” part took maybe a day or two with a few scopus searches, combing through them, finding out what’s relevant and making a coherent story out of all of this was harder and took more time. if you don’t know where even to start, maybe you should ask a professional? especially when alternative is just going in raw into the court of law, defending whatever is at stake with a few paragraphs of possibly nonsensical spicy autocomplete output
I don’t think you understand the type of multiple choice questions involved. Here’s a real question:
A father lived with his son, who was an alcoholic. When
drunk, the son often became violent and physically abused
his father. As a result, the father always lived in fear. One
night, the father heard his son on the front stoop making
loud obscene remarks. The father was certain that his son
was drunk and was terrified that he would be physically
beaten again. In his fear, he bolted the front door and took
out a revolver. When the son discovered that the door was
bolted, he kicked it down. As the son burst through the
front door, his father shot him four times in the chest, killing
him. In fact, the son was not under the influence of alcohol
or any drug and did not intend to harm his father.
At trial, the father presented the above facts and asked the
judge to instruct the jury on self-defense.
How should the judge instruct the jury with respect to
self-defense?
(A) Give the self-defense instruction, because it expresses
the defense’s theory of the case.
(B) Give the self-defense instruction, because the evidence is sufficient to raise the defense.
I’m not even going to engage in this thread cause it’s a tar pit, but I do think I have the appropriate analogy.
When taking certain exams in my CS programme you were allowed to have notes but with two restrictions:
Have to be handwritten;
have to fit on a single A4 page.
The idea was that you needed to actually put a lot of work into making it, since the entire material was obviously the size of a fucking book and not an A4 page, and you couldn’t just print/copy it from somewhere. So you really needed to distill the information and make a thought map or an index for yourself.
Compare that to an ML model that is allowed to train on data however long it wants, as long as the result is a fixed-dimension matrix with parameters that helps it answer questions with high reliability.
It’s not the same as an open book, but it’s definitely not closed book either. And the LLMs have billions of parameters in the matrix, literal gigabytes of data on their notes. The entire text of War and Peace is ~3MB for comparison. An LLM is a library of trained notes.
My question to you is how is it different than a human in this regard? I would go to class, study the material, hope to retain it, so I could then apply that knowledge on the test.
The ai is trained on the data, “hopes” to retain it, so it can apply it on the test. It’s not storing the book, so what’s the actual difference?
And if you have an answer to that, my follow up would be “what’s the effective difference?” If we stick an ai and a human in a closed room and give them a test, why does it matter the intricacies of how they are storing and recalling the data?
Me, about to suggest some actually really good, thought provoking Marvel comics that somehow got made alongside the relentless superhero soap opera: oh wait now isn’t the time, we’re dunking on the AI bro
I’m not sure what you even mean by “how is it different”, but for starters a human can actually get a good mark at the bar and spicy autocomplete clearly cannot.
What you are basing this “it clearly cannot” on? Because an early iteration of it was mediocre at it? The first ICE cars were slower than horses, I’m afraid this statement may be the equivalent of someone pointing at that and saying “cars can’t get good at going fast.”
But I specifically asked “in this regard”, referring to taking a test after previously having trained yourself on the data.
But the AI isn’t “recalling” in the same way you do, it doesn’t “remember” what it “read”, it “reads” on demand and has instant access to essentially all of the information available online it was trained on (E: though it’s becoming more or less the same thing, and is definitely the same when it comes to law books for example), from which it collects the necessary details if and when it needs it.
So yes, it is literally “sat” there with all the books open in front of it, and the ability to pinpoint a bit of information in any one of all the books in milliseconds.
Yes, it does, from the information it was trained on (or - stored), which like you say, requires a lot of hardware power so it can be accessed on demand. It isn’t just manifesting the information out of thin air, and it definitely doesn’t “remember” in the same way we do (E: even the best photographic memory isn’t the same as an indexable one).
It’s definitely not indexed, we use RAG architectures to add indexing to data stores that we want the model to have direct access to, the relevant information is injected directly in the context (prompt). This can somewhat be equated to short term memory
The rest of the information is approximated in the weights of the neural network which gives the model general knowledge and intuition…akin to long term memory
I mean you still gotta understand some shit for Ctrl+F to be helpful. If you’ve ever taken an open book quiz without prior study you’ll learn pretty quick that open book does NOT = easy A (depending on the class / prof I guess, but you get the gist).
So, open book Ctrl-F’able bar exam, I could probably get an okay score just on key word matching, not knowing jack shit about law; but it’d be far from a perfect score. Current state of machine learning appears to be in a comparable boat.
your post shows a serious lack of comprehension. just because so many of the posters in this thread are idiots didn’t mean you have to participate too.
(CPU time extremely counts, and resource-wise with these things it’s really quite a lot)
Steelmanning what this person said, I think the issue is that your ability to CTRL+F through a book during a time-limited exam is not as strong as even a single computer clocked at GHz doing the same thing. You can CTRL+F through a single book in the same time it takes it to CTRL+F through the entire body of knowledge.
that’s a misleading and meaningless way of putting it. if I rip a page out of my textbook and bring it into an exam room, I do not have with me all the data in my textbook. and yet
It doesn’t do that, either. LLMs retain the linguistic patterns found in textbooks, nothing more. It’s remarkable that they can do so much with this information alone, but it’s still a far cry from genuine intelligence.
The word parameters here must be defined. Is it the weight they are talking about or the input being used to answer the question? For the former, yeah, it’s like a person was reading a book and not an open book at all. But if it were used in the input, then it is practically an open book. They have the context on the same input.
Why is that a criticism? This is how it works for humans too: we study, we learn the stuff, and then try to recall it during tests. We’ve been trained on the data too, for neither a human nor an ai would be able to do well on the test without learning it first.
This is part of what makes ai so “scary” that it can basically know so much.
You ever meet an ai researcher with a background in biology? I’ve discussed this stuff with one. She disagrees with Turing about machines thinking including when ai is in the picture. They process information very differently from how biology does
so to summarize, your only contributions to this thread are to go “well uh you just don’t know how LLMs work” while providing absolutely no detail of your own, and reporting our regulars for “Civility” when they rightly called you out for being a fucking idiot who’s way out of their depth
Well… I do agree with you but human brains are basically big prediction engines that use lookup tables, experience, to navigate around life. Obviously a super simplification, and LLMs are nowhere near humans, but it is quite a step in the direction.
oh my, you’re such a confluence of bad takes (racist, transphobic, creepy and ignorant of the technical and biological topics you’re pontificating about.)
Because a machine that “forgets” stuff it reads seems rather useless… considering it was a multiple choice style exam and, as a machine, Chat GPT had the book entirely memorized, it should have scored perfect almost all the time.
I guess it comes down to a philosophical question as to what “know” actually means.
But from my perspective is that it certainly knows some things. It knows how to determine what I’m asking, and it clearly knows how to formulate a response by stitching together information. Is it perfect? No. But neither are humans, we mistakenly believe we know things all the time, and miscommunications are quite common.
But this is why I asked the follow up question…what’s the effective difference? Don’t get me wrong, they clearly have a lot of flaws right now. But my 8 year old had a lot of flaws too, and I assume both will get better with age.
Yeah, it’s a philosophical question, which means you need a philosophical answer. Spitballing won’t help you figure shit out a priori because it turns out that learning how to think a priorieffectively takes years of hard graft and is called “studying philosophy”. You should be asking people like me what “know” means in this context and what distinguishes memory in human beings from “memory” in an LLM (a great deal, as it happens!)
“(…) perception, attention, thought, imagination, intelligence, comprehension, the formation of knowledge, memory and working memory, judgment and evaluation, reasoning and computation, problem-solving and decision-making (…)”
From Re-evaluating GPT-4’s bar exam performance (linked in the article):
Ohhh, that is sneaky!
What I find delightful about this is that I already wasn’t impressed! Because, as the paper goes on to say
And here I was thinking it not getting a perfect score on multiple-choice questions was already damning. But apparently it doesn’t even get a particularly good score!
officially Not The Worst™, so clearly AI is going to take over law and governments any day now
also. what the hell is going on in that other reply thread. just a parade of people incorrecting each other going “LLM’s don’t work like [bad analogy], they work like [even worse analogy]”. did we hit too many buzzwords?
But LLM’s don’t work like Typewriters, they work like Microwaves!
“Nooo you don’t get it, LLMs are supposed to be shit”
Not the worst? 48th percentile is basically “average lawyer”. I don’t need a Supreme Court lawyer to argue my parking ticket. And if you train the LLM with specific case law and use RAG can get much better.
In a worst case scenario if my local lawyer can use AI to generate a letter and just quickly go through it to make sure it didn’t hallucinate, they can process more clients, offer faster service and cheaper prices. Maybe not a revolution but still a win.
good thing all of law is just answering multiple-choice tests
because judges looooove reading AI garbage and will definitely be willing to work with someone who is just repeatedly stuffing legal-sounding keywords into google docs and mashing “generate”
“guys our keyword-stuffing techniques aren’t working, we need a system to stuff EVEN MORE KEYWORDS into the keyword reassembler”
oh i would love to read those court documents
wow, negative time saved! okay so your lawyer has to read and parse several paragraphs of statistical word salad, scrap 80+% of it because it’s legalese-flavored gobbledygook, and then try to write around and reformat the remaining 20% into something that’s syntactically and legally coherent – you know, the thing their profession is literally on the line for. good idea
what promptfondlers continuously seem to fail to understand is that verification is the hard step. literally anyone on the planet can write a legal letter if they don’t care about its quality or the ramifications of sending it to a judge in their criminal defense trial. part of being a lawyer is being able to tell actual legal arguments from bullshit, and when you hire an attorney, that is the skill you are paying for. not how many paragraphs of bullshit they can spit out per minute
“but the line is going up!! see?! sure we’re constantly losing cases and/or getting them thrown out because we’re spamming documents full of nonsense at the court clerk, but we’re doing it so quickly!!”
Spoken like someone who hasn’t gotten beyond ChatGPT on default settings.
what the fuck kind of reply is this
A reply to a rant based on false premises.
congratulations on your impending graduation (to expat poster)
it’s funny how your first choice of insult is accusing me of not being deep enough into llm garbage. like, uh, yeah, why would i be
but also how dare you – i’ll have you know i only choose the most finely-tuned, artisinally-crafted models for my lawyering and/or furry erotic roleplaying needs
It’s a good thing people are so good at vigilance tasks and don’t tend to fall onto just relying on the automation.
at which point it’s just easier to do the right thing straight away, that is pay a lawyer to do their job https://www.bbc.com/news/world-us-canada-65735769
You understand that getting a list of sources and checking them is easier than finding them on your own, right?
Of course it’s even easier not checking them at all and submitting garbage, but one should have learned in 3rd grade not to submit copy-pastes from Wikipedia or any website.
This one is on human stupidity, not artifical intelligence.
so your process of getting legal advice is:
how does that simplify anything
Look it’s a really cheap and fast way of going from potential lawsuit to actual damages! That’s progress, that is!
[ed note: since I can’t markup-joke it in a way that survives lemmy: to be read in pratchett voice)
deleted by creator
that’s one weirdass assumption. when you know what are you looking for, the opposite is true. few months back i’ve authored a review chapter in my (very narrow) field, and while “getting a list of sources” part took maybe a day or two with a few scopus searches, combing through them, finding out what’s relevant and making a coherent story out of all of this was harder and took more time. if you don’t know where even to start, maybe you should ask a professional? especially when alternative is just going in raw into the court of law, defending whatever is at stake with a few paragraphs of possibly nonsensical spicy autocomplete output
I don’t think you understand the type of multiple choice questions involved. Here’s a real question:
Memorizing the book itself doesn’t teach how to answer this type of question. It requires actual application of concepts to the new facts being given.
Lol, it literally takes 5s to search for this, come up with the book and read
as the first sentence. How could you claim having that memorised wouldn’t help?
That’s like saying a person reading a book before a quiz is doing it open book because they have the memory of reading that book.
I’m not even going to engage in this thread cause it’s a tar pit, but I do think I have the appropriate analogy.
When taking certain exams in my CS programme you were allowed to have notes but with two restrictions:
The idea was that you needed to actually put a lot of work into making it, since the entire material was obviously the size of a fucking book and not an A4 page, and you couldn’t just print/copy it from somewhere. So you really needed to distill the information and make a thought map or an index for yourself.
Compare that to an ML model that is allowed to train on data however long it wants, as long as the result is a fixed-dimension matrix with parameters that helps it answer questions with high reliability.
It’s not the same as an open book, but it’s definitely not closed book either. And the LLMs have billions of parameters in the matrix, literal gigabytes of data on their notes. The entire text of War and Peace is ~3MB for comparison. An LLM is a library of trained notes.
My question to you is how is it different than a human in this regard? I would go to class, study the material, hope to retain it, so I could then apply that knowledge on the test.
The ai is trained on the data, “hopes” to retain it, so it can apply it on the test. It’s not storing the book, so what’s the actual difference?
And if you have an answer to that, my follow up would be “what’s the effective difference?” If we stick an ai and a human in a closed room and give them a test, why does it matter the intricacies of how they are storing and recalling the data?
holy fuck you’re a moron
please go read a book, and look at some art. no, marvel media doesn’t count.
Me, about to suggest some actually really good, thought provoking Marvel comics that somehow got made alongside the relentless superhero soap opera: oh wait now isn’t the time, we’re dunking on the AI bro
I’m not sure what you even mean by “how is it different”, but for starters a human can actually get a good mark at the bar and spicy autocomplete clearly cannot.
What you are basing this “it clearly cannot” on? Because an early iteration of it was mediocre at it? The first ICE cars were slower than horses, I’m afraid this statement may be the equivalent of someone pointing at that and saying “cars can’t get good at going fast.”
But I specifically asked “in this regard”, referring to taking a test after previously having trained yourself on the data.
I asked Gemini and it told me that ChatGPT can’t do shit, I’m not gonna question it.
So, it’s either perfect right now, or never capable of anything. Great critical and nuanced thinking.
Proceeds to actively engage in the thread multiple times
I never claimed to be good at self-restraint okay, everyone has their vices
You’re acting as if you never ate a full bar of chocolate after you told yourself you wouldn’t
amazed that that one reply was what they felt their contribution here had to be
It’s more like taking a digital copy into the test room with you and Ctrl+F’ing every question/answer.
Except it’s not, because they can’t perfectly recall everything.
It’s more like reading every book in the world, and someone asking you what comes next after “And I…”.
“will alwaaays love you…”
Easy. No other answer.
But the AI isn’t “recalling” in the same way you do, it doesn’t “remember” what it “read”, it “reads” on demand and has instant access to essentially all of the information
available onlineit was trained on (E: though it’s becoming more or less the same thing, and is definitely the same when it comes to law books for example), from which it collects the necessary details if and when it needs it.So yes, it is literally “sat” there with all the books open in front of it, and the ability to pinpoint a bit of information in any one of all the books in milliseconds.
It doesn’t read on demand, it reads once when it’s being trained, and it later recalls what it learnt from that training.
Training LLMs takes a very long time and a lot of hardware power.
Yes, it does, from the information it was trained on (or - stored), which like you say, requires a lot of hardware power so it can be accessed on demand. It isn’t just manifesting the information out of thin air, and it definitely doesn’t “remember” in the same way we do (E: even the best photographic memory isn’t the same as an indexable one).
It’s definitely not indexed, we use RAG architectures to add indexing to data stores that we want the model to have direct access to, the relevant information is injected directly in the context (prompt). This can somewhat be equated to short term memory
The rest of the information is approximated in the weights of the neural network which gives the model general knowledge and intuition…akin to long term memory
I mean you still gotta understand some shit for Ctrl+F to be helpful. If you’ve ever taken an open book quiz without prior study you’ll learn pretty quick that open book does NOT = easy A (depending on the class / prof I guess, but you get the gist).
So, open book Ctrl-F’able bar exam, I could probably get an okay score just on key word matching, not knowing jack shit about law; but it’d be far from a perfect score. Current state of machine learning appears to be in a comparable boat.
This is a computer. Time (in this aspect) isn’t an issue.
your post shows a serious lack of comprehension. just because so many of the posters in this thread are idiots didn’t mean you have to participate too.
(CPU time extremely counts, and resource-wise with these things it’s really quite a lot)
Steelmanning what this person said, I think the issue is that your ability to CTRL+F through a book during a time-limited exam is not as strong as even a single computer clocked at GHz doing the same thing. You can CTRL+F through a single book in the same time it takes it to CTRL+F through the entire body of knowledge.
I’m not a big AI guy but it’s really not quite like that, models do NOT contain all the data they were trained on.
Edit: I have no idea what’s going on down below this comment
lol. at least you’re honest about it
we can tell
The guy above you is right though. So what are you on about?
that’s a misleading and meaningless way of putting it. if I rip a page out of my textbook and bring it into an exam room, I do not have with me all the data in my textbook. and yet
It doesn’t do that, either. LLMs retain the linguistic patterns found in textbooks, nothing more. It’s remarkable that they can do so much with this information alone, but it’s still a far cry from genuine intelligence.
The word parameters here must be defined. Is it the weight they are talking about or the input being used to answer the question? For the former, yeah, it’s like a person was reading a book and not an open book at all. But if it were used in the input, then it is practically an open book. They have the context on the same input.
Why is that a criticism? This is how it works for humans too: we study, we learn the stuff, and then try to recall it during tests. We’ve been trained on the data too, for neither a human nor an ai would be able to do well on the test without learning it first.
This is part of what makes ai so “scary” that it can basically know so much.
Dont anthropomorphise. There is quite the difference between a human and an advanced lookuptable.
I absolutely agree. However, if you think the LLMs are just fancy LUTs, then I strongly disagree. Unless, of course, we are also just fancy LUTs.
You ever meet an ai researcher with a background in biology? I’ve discussed this stuff with one. She disagrees with Turing about machines thinking including when ai is in the picture. They process information very differently from how biology does
This is a vague non answer, although I agree it’s done very differently because our process is biological and ai is not.
But as I asked elsewhere, what’s the effective difference?
so to summarize, your only contributions to this thread are to go “well uh you just don’t know how LLMs work” while providing absolutely no detail of your own, and reporting our regulars for “Civility” when they rightly called you out for being a fucking idiot who’s way out of their depth
how fucking embarrassing for you
ah, a civility connoisseur.
Every fucking time
deleted by creator
Well… I do agree with you but human brains are basically big prediction engines that use lookup tables, experience, to navigate around life. Obviously a super simplification, and LLMs are nowhere near humans, but it is quite a step in the direction.
@phoenixz @Soyweiser “Let’s redefine what it means to be human, so we can say the LLM is human” have you bumped your head?
Well, I don’t think humans are turing computable, so I disagree with you there. Which should have been clear from my initial post.
Fun fact, what you just said about how humans are just computers is also part of dianetics. Amazing how it happened in 2 different cults.
oh my, you’re such a confluence of bad takes (racist, transphobic, creepy and ignorant of the technical and biological topics you’re pontificating about.)
I dug into their post history and god you’re fucking right
Because a machine that “forgets” stuff it reads seems rather useless… considering it was a multiple choice style exam and, as a machine, Chat GPT had the book entirely memorized, it should have scored perfect almost all the time.
I feel like this exposes a fundamental misunderstanding of how LLMs are trained.
LLMs know nothing. literally. they cannot.
Yeah but neither did Socrates
but he at least was smug about it
I guess it comes down to a philosophical question as to what “know” actually means.
But from my perspective is that it certainly knows some things. It knows how to determine what I’m asking, and it clearly knows how to formulate a response by stitching together information. Is it perfect? No. But neither are humans, we mistakenly believe we know things all the time, and miscommunications are quite common.
But this is why I asked the follow up question…what’s the effective difference? Don’t get me wrong, they clearly have a lot of flaws right now. But my 8 year old had a lot of flaws too, and I assume both will get better with age.
Yeah, it’s a philosophical question, which means you need a philosophical answer. Spitballing won’t help you figure shit out a priori because it turns out that learning how to think a priori effectively takes years of hard graft and is called “studying philosophy”. You should be asking people like me what “know” means in this context and what distinguishes memory in human beings from “memory” in an LLM (a great deal, as it happens!)
don’t compare your child to a chatbot wtf
The dehumanization that happens just because people think LLMs are impressive (they are, just not that impressive) is insane.
need to be able to think LLM’s are impressive, probably
surely tech will save us all, right?
no, it doesn’t, and it’s not a philosophical question (and neither is this a question of philosophy).
the software simply has no cognitive capabilities.
I’m not sure I agree, but then it goes to my second question:
What’s the effective difference?
don’t know why you got downvoted, an LLM is essentially a chinese room, and whether such a room “knows” is still the question.
no, it fucking isn’t. (see the postscript in linked article.)