Comment by kozzion
2 days ago
As a CTO of an Ai first law firm I would like to add that most human lawyers are absolutely terrible. We review 100s of old contracts every week and most of them have 10-20 serious drafting errors on them. The most common ones are of the order of defining the same concept like 'a working day' in various conflicting ways without every specifying when which of the conflicting definitions should be used. Lawyers still have something to add to process but for every time they catch AI out on them, the machine is usually already 10-1 ahead.
> As a CTO of an Ai first law firm I would like to add that most human lawyers are absolutely terrible
This is many people at many jobs. We continue to see this weird thing when AI comes into an industry that suddenly the product being produced by the people was perfect/amazing/whatever. Maybe that's true for the majority of HN who are lucky enough to work with experts in every field, but in the general population that's simply not the case.
Many doctors, lawyers, and programmers are not very good at their jobs - I've seen it first hand. AI gives the general population a way to steer around these people a bit, and maybe ask the right questions. Is it perfect? Nope, but it's often better than the people someone has access to.
The problem is that a layman has no way to know if what the llm told them is better than experts around them, or even just remotely correct. A wrong response and a correct, informative, helpful response look exactly the same, the LLM will express the same level of confidence and will defend them in both cases.
So, on one hand you can get an expert take that might be wrong, but is linked to an actual person, with a reputation and some level of ownership. On the other hand you have an over confident LLM that might be wrong and has no reputation, no ownership. How does that actually improve things compared to the older status quo?
Your argument that ownership and reputation generally ensures better answers than agents is absurd.
It is entirely clear that is it not the case when I compare agent generated apps with what contracted software teams have produced.
For medicine it is likely worse.
Doctors do care, but they have to give an advice based on their 15 year old knowledge - they simply can't read through 83 papers in a quick session.
I think it is a matter of time before we see the first insurance companies assign greater risk to human advice (legal, tech, medicine, etc.) than to agentic advice.
4 replies →
This argument can be applied to anything LLM generated, and doesn’t pan out. An LLM can quickly respond with cases, interpretations, and standards in minutes as a defense to its claims. You can then independently research online if you believe those claims to be correct. With a lawyer you’re lucky to get a response back in less than a week and to have your name spelled correctly on the documents they return to you.
1 reply →
> a reputation, some level of ownership, and professional insurance
ftfy
Remember the bell curve in grading? Yeah, most of the graduates from every school everywhere did NOT get straight A’s in their classes. On average, they are B students… at best. And half of them are worse than average.
> most human lawyers are absolutely terrible.
I think this extends to the other areas that AI is also disrupting.
Software engineers are also absolutely terrible: In being sloppy with handling errors, doing coercions, etc. All because always engineering after best practice simply takes too much time and mental load.
The real disruption from AI is likely that it does not get mentally fatigue. It can keep on, launch adversarial audits, make sure that everything is up to all standards.
> The real disruption from AI is likely that it does not get mentally fatigue
It's worse than mental fatigue. A friend had GLM 5.2 suddenly become dumb for hours. It started forgetting instructions, randomly inserted some bits in simplified chinese... Apparently throttled due to demand. there was tons of tokens left. (btw when your llm runs out of tokens is another example that's worse than fatigue)
A tired programmer usually knows he's tired (and can take a break) and fucks up in more expected ways
> A tired programmer usually knows he's tired
6 months ago everybody talked about how AI was not able to plan their work. They could only do minor changes.
The introspective property of knowing when you are "tired" and acting appropriately will likely be trivially observed in 6 months.
3 replies →
But the real question is 1) whether other human lawyers would catch the error 2) in how much time 3) which of the errors introduced by AI vs. human would have bigger consequences. On 1 and 2 the answers are very likely to be "yes" and "it would take humans more time", but the question that really matters is 3.
Adversarial agents and reviews means 3 is very clearly giving the benefit to AI.
Let's face it, only thing lawyers are there for right now is to take liability. And even that may not be needed 5 years from now.
Middlemen that can be automated will be automated.
Lawyers do a lot more than accepting liability. In my field they spend a lot of time on contract negotiation, and on interpreting what federal and state regulations actually require us to do.
I have the impression that most law firms consume and produce documentation by means of a cadre of paralegals who are minimally trained, minimally experienced, and maximally stressed by the pressure of deadlines and case loads. These paralegals are probably working to read hundreds of pages of legal filings and make sense out of them, and on the other end, they are working to produce as much documentation as possible and that's going to be by cut-and-paste assembly. So no, they are not going to understand stuff very well, and no, their output is not really going to be consistent or intelligible. But the managing attorneys depend on these paralegals to get anything done, just like Congresspeople rely on their staffers to actually read bills and draft legislation.
At the end of the day, I think these paralegals have been quite weak links, and nearly nobody will notice a difference in quality when it's all handed over to LLMs. But the paralegals will lose their jobs, and probably have nothing better to do than purchase angle grinders and cut down Flock camera poles, or crown the front hood of Waymos with a traffic pylon.
It's really the junior associates doing most of that grunt work. Paralegals in most firms have more limited responsibilities. Some work has already been handed over to LLMs but the job market for graduates of first-tier law schools remains strong.
Wonderful, I'm tired of lawyers. Happy to see that evolve and figure out how to provide actual value for its cost.
> most human lawyers are absolutely terrible. We review 100s of old contracts every week
I hope the irony is lost on no one that "Attention Is All You Need" was the paper that kicked the LLM boom off.
Part of the distinction between a how good one lawyer will be vs another is exactly that: attention to detail. And, as a trade, a practice, it seems like a skill that they try to hammer into the people who pursue this career (to varying degrees of success clearly).
It makes the last 20 years of leet code hiring look downright wrong. Your ability to recall what sorting algo to use, or solve some brain teaser conveys nothing about your ability to read (massive amounts) of code, and think critically about it.
Most people are terrible at what they do, period, regardless of profession.