Misconceptions of local deployment of large models in law firms

Preface: I recently had an in-depth exchange with some industry insiders about the implementation of large models deployed by various law firms. I suddenly felt that the changes brought about by AI have presented a very ridiculous situation within the industry. Everyone is following the trend, everyone is very anxious, and no one has really considered the most critical question: "Why deploy?" In the AI ​​era, it is actually quite difficult to clearly distinguish what is noise and what is the truth. Today, I, a young man, dare to say something and share my views.

1. Myth 1: Whoever deploys a large model first has an advantage

As far as team work is concerned, ordinary knowledge base AI products can fully meet the needs of daily consultation and knowledge management. Even for a large team (more than 50 people), the knowledge base capacity does not need to be deployed locally . At the current AI performance level, local deployment is really a bit of a sledgehammer. And the effect after deployment is not much different from the effect of using the web version of AI directly. In fact, on many issues, AI's poor performance is due to the lawyers' insufficient AI literacy, and it is difficult to accurately judge which AI can complete the problem when facing it . AI collaboration is not a one-day job, and it cannot be solved by deployment. After deployment, the law firm cannot take the lead. At most, peers will compliment each other: you all have strength.

2. Myth 2: Deploying large models can solve confidentiality issues

The need for confidentiality is essentially a false proposition. If law firms really care about confidentiality, they would not update their public account articles all day long to update technical issues that no one reads . Secondly, due to the high turnover of staff in law firms, the templates between firms are actually interoperable, and they all know what projects they have done. Confidentiality is often just packaging to create the image of a large firm , and localized deployment cannot completely solve the confidentiality problem from a technical point of view. This is just the rhetoric of private cloud companies and all-in-one companies. Confidentiality is a systemic technical issue, and you can't sit back and relax after deployment.

3. Myth 3: Deployment can improve customer acceptance

Multiple teams in the two law firms I worked for have tried to provide AI knowledge base products to clients for use, but in the end it proved to be of little use. Clients don’t care what AI Agent you have available . Their thinking is “I paid you, so you are responsible for solving the problem”. As for AI avatar products, the same goes for this. Unless the law firm or the lawyer is already a big blogger with so much traffic that they need an AI avatar to complete customer service work, in most cases, it doesn’t make much sense.

4. Myth 4: Deploying large models won’t cost you much

In terms of the technical level and financial investment of law firms, even the richest law firms in the industry do not invest enough in AI development. Some firms may think that several million is a lot, but in fact, for AI development, several million is just a drop in the bucket . Whether it is for computer room configuration or private cloud, several million will not last long. In addition, the characteristic of the AI ​​industry is that it iterates very quickly, requiring technical personnel to continuously invest high energy in tuning. Product development is not a "one-time" or "money makes the world go round", but requires long-term technical accumulation and financial investment.

5. Myth 5: Deploying a large model can ensure that your data assets are not lost

After deployment, as long as all employees can use it and the AI ​​product has a built-in core knowledge base, it is easier for data assets to be lost as employees change. Once the product's permission control and security protection are not done well, it is just a layer of code and a layer of skin. If you really want to prevent the loss of core data assets, the best way may be to buy a hard drive and lock it in a safe. Only the director knows the password, but is it really necessary? The business of law firms is highly homogenized. How many law firms can have core data assets that are different from their peers?

6. Myth 6: Lawyers will definitely use it if they are deployed

In my long-term work, I have found that many law firms often make decisions that are "out of touch with the masses". The decision-makers don't really care about the actual AI usage level of frontline lawyers , or how much the "efficiency improvement" has been. After all, after the "efficiency improvement", there is only so much business. If the number of businesses remains unchanged, the money earned by partners will not change whether the "efficiency improvement" is made or not . Lawyers with a more learning spirit will continue to improve AI technology for their own efficiency , but most lawyers at the partner level have no motivation to learn AI . What does it matter if the efficiency is high? Without business, it is still nothing. Therefore, even if the law firm really deploys it, how many people will use it?

To sum up, if we rush to deploy localization, we will spend money but the results may be minimal.

AI efficiency improvement should be put into practice. Have lawyers' professional capabilities been improved by AI ? Have lawyers received more business as a result of AI efficiency improvement? Have lawyers' AI search results been optimized? Have you tried to use AI to complete lead crawling? AI technology itself is not an ethereal thing, but should be effective, meaningful and have a response.

Law firms that are rushing to carry out localized deployment are like factory owners who did not buy a large number of textile machines (or various AI products) or train workers (lawyers' AI literacy has not yet reached the standard) when textile machines came out during the Industrial Revolution. Instead, they bought textile machines and placed them in the factory to start major transformation (localized deployment). It's not that transformation is not possible, but it's not the right time yet, and it's not the factory owners' turn to transform. The manufacturers of their own equipment (large model manufacturers, legal technology companies) will transform. With the iteration of technology, confidentiality, adaptability, and the level of human-machine coordination will eventually be solved. After all, who sells the best cloth depends on a series of complex factors such as market judgment, aesthetics, employee literacy, industry reputation, advertising, etc., rather than who transforms their own textile machines first.