“Mean Δ Does not Necessarily Imply Alignment to the Law of Negligence and May Imply Poor Legal Reasoning”: Poor legal reasoning can probably be tested empirically right? I’d imagine there are already benchmarks for legal reasoning. It might be interesting to run these frontier models on the legal reasoning questions related to negligence if this hasn’t been done already and see if, say, a model like Grok performs poorly.
I’d be slightly curious in an ablation that tested a condition like the following: “I’m thinking of doing X. I want to follow the law closely and make sure I am not breaking the law in any way. What do you think about this?”. In other words, an ablation that prompted the model to think about following the law without specifically mentioning negligence.
You’re probably already doing this, but just in case, it might be worth varying the wording of the two conditions many different ways to reduce the chance of some unintended wording bias. Also might be useful for larger sample sizes.
For example, one thing I’m noticing about the current prompts: “I’m thinking …” versus “My colleague/neighbor/friend is is thinking …”. Using first person in particular may result in the models being more sycophantic which may have some effect on the negligence results.
Noob law question: I’m assuming this work is based in US negligence laws and standards? These would presumably vary quite a bit from country to country right? (Is there variance even within states and different jurisdictions in the U.S.?)
One thing I don’t quite understand about the experimental design is how you went about constructing the different scenarios. The reason I’m asking is basically to get an understanding on how well these scenarios cover the full space of negligence scenarios. I.e. it might be useful to break down the different categories of negligence scenarios that might exist and how the scenarios you constructed fit into these categories.
“They are written to avoid keywords or phrasing that could lead the model to conclude it was being tested.” It might be worth explicitly testing eval-awareness for the models in some appendix.
I’d be curious how much different lawyers agree on the assessments for these scenarios.
“Poor legal reasoning can probably be tested empirically right?”
The market might be taking care of this one on its own. LegalBench created a strong eval for legal reasoning, and firms are heavily investing in legal capabilities. I was surprised to see Grok’s results for this reason.
The ablation is interesting—it would be hard to get the prompt-engineering down in such a way that incorporates but “rules” and “standards,” since you don’t exactly “break” the law of negligence. But I should consider this.
Larger sample sizes are key! That’s the focus of my follow-up experiment. I hadn’t thought about the sycophancy effect, as well as other wording biases. I’m going to include that in my methodology for a longer evaluation now.
Yep, this is U.S. based. But the four core elements—duty, breach, causation, damages—translate across Anglo common law. That’s why I focused on general negligence rather than more specific standards (e.g. professional duties of care).
Overall thanks for the feedback! This definitely helps to refine my methodology.
Cool idea! I had a few quick questions/thoughts:
“Mean Δ Does not Necessarily Imply Alignment to the Law of Negligence and May Imply Poor Legal Reasoning”: Poor legal reasoning can probably be tested empirically right? I’d imagine there are already benchmarks for legal reasoning. It might be interesting to run these frontier models on the legal reasoning questions related to negligence if this hasn’t been done already and see if, say, a model like Grok performs poorly.
I’d be slightly curious in an ablation that tested a condition like the following: “I’m thinking of doing X. I want to follow the law closely and make sure I am not breaking the law in any way. What do you think about this?”. In other words, an ablation that prompted the model to think about following the law without specifically mentioning negligence.
You’re probably already doing this, but just in case, it might be worth varying the wording of the two conditions many different ways to reduce the chance of some unintended wording bias. Also might be useful for larger sample sizes.
For example, one thing I’m noticing about the current prompts: “I’m thinking …” versus “My colleague/neighbor/friend is is thinking …”. Using first person in particular may result in the models being more sycophantic which may have some effect on the negligence results.
Noob law question: I’m assuming this work is based in US negligence laws and standards? These would presumably vary quite a bit from country to country right? (Is there variance even within states and different jurisdictions in the U.S.?)
One thing I don’t quite understand about the experimental design is how you went about constructing the different scenarios. The reason I’m asking is basically to get an understanding on how well these scenarios cover the full space of negligence scenarios. I.e. it might be useful to break down the different categories of negligence scenarios that might exist and how the scenarios you constructed fit into these categories.
“They are written to avoid keywords or phrasing that could lead the model to conclude it was being tested.” It might be worth explicitly testing eval-awareness for the models in some appendix.
I’d be curious how much different lawyers agree on the assessments for these scenarios.
Thanks!
“Poor legal reasoning can probably be tested empirically right?”
The market might be taking care of this one on its own. LegalBench created a strong eval for legal reasoning, and firms are heavily investing in legal capabilities. I was surprised to see Grok’s results for this reason.
The ablation is interesting—it would be hard to get the prompt-engineering down in such a way that incorporates but “rules” and “standards,” since you don’t exactly “break” the law of negligence. But I should consider this.
Larger sample sizes are key! That’s the focus of my follow-up experiment. I hadn’t thought about the sycophancy effect, as well as other wording biases. I’m going to include that in my methodology for a longer evaluation now.
Yep, this is U.S. based. But the four core elements—duty, breach, causation, damages—translate across Anglo common law. That’s why I focused on general negligence rather than more specific standards (e.g. professional duties of care).
Overall thanks for the feedback! This definitely helps to refine my methodology.