> AI can be used for research on papers ONLY for the limited purpose of identifying sources, such as cases, statutes, or secondary sources. Students are responsible for the accuracy of their research and all other aspects of their submitted work.
Absolutely level-headed policy. Whoever came up with this deserves some serious praise.
My understanding is that this is analogous to AI use policies within many law firms. Using AI-powered search tools as part of research is usually allowed; using generative AI to draft or rewrite content is usually not.
(Given the importance of confidentiality, law firms also tend to be pretty picky about which tools their staff use.)
Maybe this is obvious or overly reductive, but that rings well for me too and sounds so much like early (and still wise) guidance for Wikipedia and maybe the Internet at large.
I’m curious how they draw the line for spelling and grammar checking. I’ve never seen a law school (or other graduate program) prohibit standard spell-checking or grammar-checking in software.
No judge has ever sanctioned a lawyer because their spell checker wrote a whole hallucinated document for them. So spell checkers are fine. Any other questions?
> Future lawyers may need to use artificial intelligence (“AI”) fluently. But the current state of the technology requires that AI use be coupled with the cognitive skills necessary to strategically deploy the technology, to critically assess its work product, and to uphold ethical obligations to clients and to the legal system.
I think I can understand the reasoning behind most of those rules, but the prohibition against AI translation of papers not in English is a bit puzzling. Do they expect students to do their own translations in all cases?
Banning Ai outside of the classroom is impossible, like telling students they cannot use google for take-home tests. Students will always find ways to circumvent these restrictions.
> AI use is prohibited for any use for any purpose in any exam situation.
Read literally this bans the use of hearing aids in an oral examination. All the devices on the market today use noise reduction based on neural networks or transformers. Isn't that AI? (The manufacturers' ad copy certainly insists so.) Presumably not what they intended but "AI" is not defined in the document.
ADA accommodations regularly change what can and cannot be used during an exam. This is not the “gotcha” you think it is. Any lawyer making an argument along these lines probably used AI to pass the bar.
For example, a blind person could still get an accommodation to use a AI vision tool which reads the exam to them. Or the school could give them a human who reads the exam to them. Either would meet the ADA accommodation requirements, and neither would actually be blocked by UC Berkeley’s new “no AI” rule because the ADA supersedes this “no AI” rule.
These days there’s often a tiny neural net inside good hearing aids which decides which frequencies to amplify by how much. That way they can dynamically detect and remove background noise and isolate actual speech. Among many other features.
All of the major hearing aid manufacturers incorporate DNNs or transformers into their chipsets. The manufacturers advertise it as "AI" and that's probably a fair description.
> > Our goal was to train the DNN on sound scenes so that it could solve the task of balancing sound sources by preserving cues and attenuating noise. The large amount of data needed for this training was recorded in different sound scenes across a wide array of listening environments representing sound scenes that listeners would typically be exposed to in their everyday lives. We used a specialised spherical microphone, capable of capturing 360 degrees of sounds to provide the DNN with a spatially accurate and detailed sound scene and to train it on the full sound scene.
They claim it improves signal to noise over traditional DSP hearing aids.
There's a pretty wide difference between what we call AI and what salespeople call AI. Technically, a linear regression can be modelled as a single node neural network, which makes it AI.
It'd be pretty typical of sales to try to make hearing aids sound like they're actually capable of thought comparable to people. AI is machines that can produce results approaching thought, and I don't believe these hearing aids are going to be replacing any brains any time soon. Claude, on the other hand... It's already replaced a large chunk of people at their jobs.
Wouldn't also any sufficiently complicated expert system be AI? Now how many if blocks you need for it to be sufficiently complicated might be vague...
Law is so ripe for replacement by AI. It is entirely based on established written arguments with gray areas, edge cases, linguistic subtleties, etc., being the crux upon which many controversial decisions are drawn. That is, law arguments are based on a mastery of language coupled with knowledge of legal precedent. LLMs are extremely good at linguistic reasoning and can be trained on ever legal case every published.
You seem to have the impression that judges follow the law and see where it leads. In reality, judges tend to decide where they want to go and then figure out how to get there.
Judging and litigation will be among the last areas to be replaced by so-called AI, if indeed they ever are. People are infinitely complex and tend to disagree with one another.
Those in power tend not to want to give up their power. And judges are where they are because they want to be in power. So my guess is that end point will be entirely arbitrary. AI is used to get decision that judge wants one way or an other.
> AI can be used for research on papers ONLY for the limited purpose of identifying sources, such as cases, statutes, or secondary sources. Students are responsible for the accuracy of their research and all other aspects of their submitted work.
Absolutely level-headed policy. Whoever came up with this deserves some serious praise.
My understanding is that this is analogous to AI use policies within many law firms. Using AI-powered search tools as part of research is usually allowed; using generative AI to draft or rewrite content is usually not.
(Given the importance of confidentiality, law firms also tend to be pretty picky about which tools their staff use.)
I've been running deer-flow (https://github.com/bytedance/deer-flow) for doing research and it's quite good even with low context sizes.
The irony of not being able to use LLMs for an LLM degree
Maybe this is obvious or overly reductive, but that rings well for me too and sounds so much like early (and still wise) guidance for Wikipedia and maybe the Internet at large.
University of Chicago's policy is more interesting: https://www.law.uchicago.edu/news/ai-strategy-statement
Earlier discussion of the University of Chicago Law School's policy: https://news.ycombinator.com/item?id=49024980
I’m curious how they draw the line for spelling and grammar checking. I’ve never seen a law school (or other graduate program) prohibit standard spell-checking or grammar-checking in software.
No judge has ever sanctioned a lawyer because their spell checker wrote a whole hallucinated document for them. So spell checkers are fine. Any other questions?
MS Word uses ML in grammar and spell checking.
Related patent: https://image-ppubs.uspto.gov/dirsearch-public/print/downloa...
As long as you are using red squiggles instead of blue you should be fine though.
IMHO though, they should say GenAI as AI is too broad.
Aren't autocorrect mistakes basically spellcheckers hallucinating?
> Future lawyers may need to use artificial intelligence (“AI”) fluently. But the current state of the technology requires that AI use be coupled with the cognitive skills necessary to strategically deploy the technology, to critically assess its work product, and to uphold ethical obligations to clients and to the legal system.
So, not today, SlopSatan
You have to be jacked to use the jackhammer.
I think I can understand the reasoning behind most of those rules, but the prohibition against AI translation of papers not in English is a bit puzzling. Do they expect students to do their own translations in all cases?
Banning Ai outside of the classroom is impossible, like telling students they cannot use google for take-home tests. Students will always find ways to circumvent these restrictions.
Most of this policy is unenforceable.
> AI use is prohibited for any use for any purpose in any exam situation.
Read literally this bans the use of hearing aids in an oral examination. All the devices on the market today use noise reduction based on neural networks or transformers. Isn't that AI? (The manufacturers' ad copy certainly insists so.) Presumably not what they intended but "AI" is not defined in the document.
ADA accommodations regularly change what can and cannot be used during an exam. This is not the “gotcha” you think it is. Any lawyer making an argument along these lines probably used AI to pass the bar.
For example, a blind person could still get an accommodation to use a AI vision tool which reads the exam to them. Or the school could give them a human who reads the exam to them. Either would meet the ADA accommodation requirements, and neither would actually be blocked by UC Berkeley’s new “no AI” rule because the ADA supersedes this “no AI” rule.
Why would you consider an amplifier AI?
These days there’s often a tiny neural net inside good hearing aids which decides which frequencies to amplify by how much. That way they can dynamically detect and remove background noise and isolate actual speech. Among many other features.
like, a band pass filter?
Do you have any example devices?
All of the major hearing aid manufacturers incorporate DNNs or transformers into their chipsets. The manufacturers advertise it as "AI" and that's probably a fair description.
It's a kind of filter, implemented with neural networks, sound-to-sound translation. For example Oticon's white paper on their AI based noise reduction system is here https://assets-je.cas.dgs.com/-/media/oticon/main/pdf/austra...
> > Our goal was to train the DNN on sound scenes so that it could solve the task of balancing sound sources by preserving cues and attenuating noise. The large amount of data needed for this training was recorded in different sound scenes across a wide array of listening environments representing sound scenes that listeners would typically be exposed to in their everyday lives. We used a specialised spherical microphone, capable of capturing 360 degrees of sounds to provide the DNN with a spatially accurate and detailed sound scene and to train it on the full sound scene.
They claim it improves signal to noise over traditional DSP hearing aids.
There's a pretty wide difference between what we call AI and what salespeople call AI. Technically, a linear regression can be modelled as a single node neural network, which makes it AI.
It'd be pretty typical of sales to try to make hearing aids sound like they're actually capable of thought comparable to people. AI is machines that can produce results approaching thought, and I don't believe these hearing aids are going to be replacing any brains any time soon. Claude, on the other hand... It's already replaced a large chunk of people at their jobs.
I consider most deep learning and statistical learning systems to be AI.
But I also consider traditional symbolic AI to be AI.
You only underline my original point: no one can define AI.
Wouldn't also any sufficiently complicated expert system be AI? Now how many if blocks you need for it to be sufficiently complicated might be vague...
Is linear regression AI?
Because the specific context here is law, it seems fair to answer: “depends on whether my client wants it to be AI or not.”
Law is so ripe for replacement by AI. It is entirely based on established written arguments with gray areas, edge cases, linguistic subtleties, etc., being the crux upon which many controversial decisions are drawn. That is, law arguments are based on a mastery of language coupled with knowledge of legal precedent. LLMs are extremely good at linguistic reasoning and can be trained on ever legal case every published.
You seem to have the impression that judges follow the law and see where it leads. In reality, judges tend to decide where they want to go and then figure out how to get there.
Judging and litigation will be among the last areas to be replaced by so-called AI, if indeed they ever are. People are infinitely complex and tend to disagree with one another.
Those in power tend not to want to give up their power. And judges are where they are because they want to be in power. So my guess is that end point will be entirely arbitrary. AI is used to get decision that judge wants one way or an other.