Lawyer Lin Shanglun's Article: Amplify and Accelerate — The AI Capabilities That 99% of Legal Professionals Missed

Over the past six months, I have assisted thousands of legal professionals worldwide in integrating AI. From lawyers in the United States, Japan, and Australia to Taiwan, as well as various NGOs and public interest groups, I have participated in over a thousand onboarding meetings. Through this opportunity, I want to thoroughly review some of the most common misconceptions about AI observed in modern times. Because I simply cannot stand seeing so many people underestimate AI due to misunderstandings, leading to attacks and resistance—it's truly a pity.

The biggest misconception: "My workflow doesn't need AI"

In my view, the most widespread and also the most fatal misconception is that many say, "My workflow probably doesn't need AI." This statement itself is a severe underestimation of AI's capabilities. And this underestimation often stems from several different causes.

The first cause is "poor usage," leading people to believe that AI outputs are terrible. Many lawyers see their clients presenting GPT-generated content and instinctively think, "This stuff is so useless." But this is a complete misunderstanding of AI. The key to AI output quality actually depends on what input you give it. The kind of input a user provides determines the quality of the results.

Imagine: your client is someone who doesn't understand law. If you mock a layperson for producing poor AI results and then conclude that you don't need AI, how absurd is that? AI is fundamentally a tool for "amplification" and "acceleration": amplifying your work results, speeding up your workflow. Someone who can't distinguish between civil law and criminal law, or doesn't know what programming law is, will naturally operate poorly and lack the ability to iterate on results each time. This is a user problem, not an AI problem.

The second cause is "using consumer-grade models without understanding their limitations"

The second cause is that users employ "consumer" models without realizing their inherent issues. Think carefully: these are consumer AI models, not enterprise-grade or commercial AI, and their design goals and billing logic are fundamentally different.

Today’s large language models are already incredibly powerful in document processing, far beyond what most people imagine. They can generate text combinations, and after iteration, produce results indistinguishable from humans—merely mimicking style; to surpass humans, they just need a few more iterations. But why do many people think the results are inadequate? The key lies in the choice of model.

Here’s a decisive difference: billing basis. Consumer models typically charge based on input tokens, and the amount of data they read per session is very limited—often only 10,000 to 20k tokens. When you use such an "all-you-can-eat" model, providers, aiming to cut costs and maximize profits, will naturally try to minimize input token usage.

What happens then? For example, if you upload a 50k-word contract or judgment, it might only read 20,000 tokens. It might only read a few paragraphs or just the first 10,000–20,000 words, ignoring the rest. As a result, you might find that:

• Your long contract misses miscellaneous clauses;
• Your judgment document's final ruling or confiscation details aren't read;
• Your lengthy meeting transcript is missing parts in the summary.

This is not an AI capability issue but a model choice error. With current technology, no matter how many documents you upload, they can be read precisely. If your billing were based on "per token," I believe providers would happily read every word you send. Once input is ensured, based on my experience of over a thousand coaching sessions, professional legal results rarely disappoint.

"I write better than AI": naive and a misjudgment

Next, many emphasize: "My writing ability surpasses AI." I must say, this mindset is very naive.

You should know that the top and most profitable law firms in the U.S. are all using AI across the board—from summer interns, senior lawyers, to partners—all have AI in their workflow. Why? Because when a task can be automated and generated by machines, no one will go back to old methods. It’s like having a calculator—you won't manually do arithmetic; taking photos to record instead of copying by hand; flying instead of walking or sailing. That’s the essence of progress.

It’s similar to: the top four most profitable law firms in the U.S. are using AI, just like this year's NBA champions acknowledge that "the basketball is very easy to play." Yet many people deny this, citing strange exceptions—like a third-world basketball team claiming "this ball is hard to play with." Such arguments are irrational and, statistically and logically, hardly convincing.

Mistaken usage: Overlooked "Loose Attention"

Besides model choice, incorrect usage methods are also common. AI is very prone to "loose attention"—scattered focus.

The most correct approach, I believe all providers know, is to first process data via a database (vectorization). If you look into how it works, it essentially converts every document into plain text, then feeds it to AI for interpretation and analysis. Why? Because AI's reading performance on plain text is optimal; PDFs are better than images, DOC files are worse, and images with text are even less effective.

Therefore, for analyzing the same data, the proper priority should be:

• Divide into multiple files, rather than combine into one;
• Combine into one file, rather than vectorize all data into text first, then let AI read.

But if you don’t know these principles, you might upload four or five files at once, then find the results unsatisfactory, and instinctively think "AI is no good." In fact, when someone has already used AI to surpass human performance, your mistake is in the usage method, not in AI's capability. The real short-sightedness is blaming AI instead of improving your approach.

AI’s strongest ability: "Infinite transformation" of format and form

I want to specifically highlight AI’s ability in "format and content transformation," which many overlook. When you have substantial and sufficient content, AI’s capacity for any form of format conversion is its strongest, most powerful feature.

This also changes how lawyers charge for services. Previously, you had to feel "it's time to bill" before doing certain tasks. Now, with AI, you can extend the billing point infinitely, greatly refining your service without additional charges. For example, a one-time consultation used to end after the session; now, after the consultation, you can have AI analyze related data, draft legal documents, and more.

Take a labor dispute case: if an employee and employer have a conflict, with the client’s consent, you can organize the entire meeting into minutes. This record might include:

• Future litigation strategies;
• How to file complaints or appeal to authorities;
• How to claim labor law Article 14 rights for forced resignation and severance;
• Calculations of severance pay, overtime, work hours, pension contributions, etc.

With such clear content, what can you transform it into? A proof of termination letter, negotiation letter, complaint, defense, or a reasoned statement, even a public announcement.

The key is: when your content is clear, even if the input contains verbose or irrelevant parts, AI can accurately read, extract, and interpret as long as you use the right concepts and tools. The input doesn’t need to be "refined" or "fixed format"; it just needs to be "clear, rich, and comprehensive." As for the format, transforming it becomes effortless with AI.

The overlooked goldmine: AI’s "Unlimited Iteration" ability

Next, I want to discuss a capability many overlook: the "infinite iteration." What is iteration? Let me illustrate with a birthday card example.

First version: You want to write a birthday card for your lover. A careless guy might just write something like "Love you forever, sweetheart," which is perfunctory and insincere. But for AI, this version only needs your lover’s name, how you love them, and your conclusion—that’s a "draft."

Second version: You tell AI, "Tell me what meaningful things have happened over the years that I should be grateful for." Even if you’re unsure how to incorporate this, just provide the facts, and AI will automatically weave these beautiful love stories into your rough draft.

Third version: You find a scene from "Pride and Prejudice" that depicts a perfect love scene, and feed this to AI for a third iteration. The resulting birthday card becomes a well-structured, historically rich, and movie-quoting masterpiece.

In the pre-AI era, making three revisions might take an hour each time; but in the AI world, one modification could have already undergone countless iterations.

A birthday card is just a small example. For lawyers, a lengthy legal brief or declaration can be improved much faster through iteration than manual revisions. Senior lawyers often criticize junior lawyers: "Junior lawyer submits one draft, I criticize and send back; second draft, criticize again, send back." That’s highly inefficient. Today, you should leverage AI’s rapid iteration to quickly improve each document’s quality. That’s the correct way to use it.

Why professional workers need "dedicated AI"

But here’s a crucial reminder: if you use traditional or incorrect models, you cannot perform proper iteration. Why? Because each iteration consumes more expensive input and output tokens, which consumer models aim to minimize.

For example, if you have a 5,000-word article divided into several parts, and you only want to modify the third part, what happens? Consumer models will directly insert "omitted" or similar placeholders in parts you didn't specify to modify, thinking "since you didn't ask for changes, I won't generate them to save tokens." But that’s wrong. In traditional human workflows, each revision is based on the previous version. To avoid wasting tokens, AI will delete potential revision areas, insert "omitted," or perform large merges, making iterative refinement impossible. Without iteration, you’re essentially giving up AI’s most powerful feature, which is a huge loss.

That’s why lawyers worldwide need dedicated AI, data workers need dedicated AI, journalists need dedicated AI, writers need dedicated AI—because each profession requires a token usage method tailored to its needs.

Consumer models are for whom? For ordinary people, for simple questions like "What should I do about my cough and runny nose?" When it involves professional content, iteration is necessary. Using consumer models for professional tasks is a fundamental misjudgment of AI’s nature.

Don’t use others’ mistakes as an excuse not to improve

Many people look at negative AI news and feel proud they haven't used AI. Recently, some U.S. law firms were sanctioned because AI produced "judgment hallucinations." Seeing such news, some share it loudly, declaring "AI must not be used."

But do you realize? For every law firm that faced such issues, how many others avoided them and thus saved time and increased productivity? These top firms have publicly disclosed their AI tools. Frankly, those that ran into trouble used the wrong models—probably consumer-grade ones without proper checks—and failed to recognize that AI’s core is to "amplify and accelerate your professional ability," not to "replace your judgment." That’s why problems happen.

Blaming others’ mistakes as a reason not to adopt AI or to avoid progress is very unfortunate and limiting. It’s completely out of sync with the current reality.

AI is a "Universal Technology," yet adoption is shockingly low

Everyone should understand that AI is a "universal technology." It’s accessible to everyone, with a very low learning curve, and highly intuitive—just like smartphones. When you use a smartphone, you might wonder: how many people still use "dumb phones"? Probably less than 10%.

But AI, being so easy to access, powerful, and user-friendly, has a global adoption rate of less than 1% or 2%. I find this extremely regrettable. From another perspective, this is the best opportunity for those who understand AI now—because when 99% of people haven't yet joined, those who get on board early can unleash astonishing productivity.

So, how come I am so confident about every lawyer’s AI usage and pain points? Because over these six months, I have truly helped nearly a thousand legal professionals worldwide adopt and understand AI, accumulating these insights.

I also believe these tools are even more vital for NGOs and public interest groups. I want to especially thank the Judicial Reform Foundation and the Taiwan Innocence Project for their recognition. Helping public groups integrate AI into their work is what I find most meaningful. These organizations handle a lot of heavy paperwork, and after integrating AI, they have greatly reduced manpower needs and achieved excellent results. I believe I’ve been of real help. I am very grateful for their trust, and it’s my honor to contribute as an AI lawyer in this era.

The Taiwan Foundation for Democracy, through a certificate of appreciation, recognizes Lawyer Shang-Lun Lin for providing free M-ROSS.AI system support, assisting with foreign language document translation, judgment summaries, and text organization.

The Taiwan Innocence Project awards a certificate of appreciation for M-ROSS.AI’s application in foreign document translation and summaries, Chinese judgment summaries, and optical character recognition of non-sensitive files.

This article is truly just because I see so many people attacking and resisting AI due to a mistaken belief. I simply cannot accept it. I must say: those who actually use AI will only continue to mock such resistance. Because this misconception actually puts you far behind your competitors, making it impossible to keep up and leaving you far in the rear of this era.

Frequently Asked Questions

What’s the difference between consumer AI and enterprise or dedicated AI?

The key lies in billing and input tokens. Consumer models aim to lower costs by reading only one or two ten-thousand tokens per session, so long contracts or judgments are prone to being read incompletely. Dedicated AI billed per token reads every word thoroughly, resulting in much higher stability and accuracy.

Why is "infinite iteration" the most valuable capability of AI?

Human work is based on repeatedly revising previous versions. AI can iterate countless times in a very short period. But consumer models, to save tokens, often mark unmodified parts as "omitted," effectively abandoning iteration. This makes professional document refinement difficult.