Hacker News Reader

Comment by kozzion | Hacker News Reader

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.

View on HN

> 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.

15 year old knowledge is fine in most cases. Human diseases and their treatments haven't really changed in that time, and those that have are pretty well known. You will get the "standard of care" as a starting point.

You are citing a legacy institutional system where updating a "standard of care" was a major institutional chance.

You likely have to change your idea about this. Heck, this view was wrong 6 months ago. Sticking with is becomes a hazard to patients.

> Your argument that ownership and reputation generally ensures better answers than agents is absurd.

That's not my argument... I didn't draw the conclusion that because of those 2 factors humans generally provide better answers. I personally have no idea if that's the case, and for sure wouldn't rely on my personal feelings to evaluate that

You clearly appear to setup that argument

> 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.

Regardless, I agree.

The newest studies still works on llms that are two years old. It doesn't appear that proper medical harnesses with frontier models has been evaluated.

My slight intuition is that we will already now see results that are much better than average human doctors.

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.

> This argument can be applied to anything LLM generated

That's the whole point, yes. LLMs are inherently unreliable. Both human experts and LLMs are unreliable in their own ways. The human expert has a reputation and some level of responsibility, the LLM doesn't

> 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.

human fatigue is detected by sensory experience and there's no sensory experience for llms. so it can just guess.

operators will never expose this guess to the user (who wants to pay for a tool that complains and refuses work?) so they probably will use it to load balance differently. and if they are out of resources then they can't load balance so they'll just hope you won't notice, so we're back to where we started. maybe it's already roughly how it works now...

what's most likely in 6 months is a big correction in pricing, once they get people hooked up on this enough. maybe there will be a model that manages to detect when it's stupid but it won't be available outside FAANG and DOD

Ahh sorry, I see that you explicitly wrote that deteriorated performance was due to throttling.

Naturally, LLMs do not get fatigue. As such they also don't need to detect it. That is correct. But LLMs can see if their performance have fatigue-like deterioration and correct for it.

That case is even more trivial and has nothing fundamentally to do with LLMs. Capacity is added everyday.

I would not base my long term projects on what happens in the market based on the fact that a single provider needs to throttle to satisfy demand now.

Your assumption that the current pricing is not sustainable might be fair. Personally I believe the opposite, and I havde not seen any indication that inference should not continue to decline in price.

In particular, LLMs appear to be hyper commoditisable. So if Anthropic or OpenAI is doing pricing shenanigans, people will likely move on.

> Naturally, LLMs do not get fatigue

they are not human so human sensations don't apply. but I heard when you run out of tokens it's also kind of like fatigue, just less predictable

> 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.

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 people are terrible at what they do, period, regardless of profession.

Reply on news.ycombinator.com