The term of the year 2023 in our firm was clearly GPT, the advanced language model from OpenAI. It saw mass adoption in November 2022 after the launch of chat.openai.com, the chatGPT application. Our law firm, ARROWS, has been working very intensively with AI implementation throughout 2023. The purpose of this article is to summarize our crucial experience and share it with those of you who think they missed something. On the contrary, no one has missed the train yet, there is still time to jump on. But the sooner the better. This article is here to help you overcome the first hurdles.
It's clear that AI not only brings new opportunities for automation and efficiency, but has also fundamentally changed the way companies operate and, more importantly, will operate very soon. We at ARROWS have an internal motto of "tradition, innovation, efficiency". This means that we value our traditional industry, deploy methods to it, and strive to be as cost-effective as possible for our clients. Do you have a similar approach in your company? Then read on - it's a good fit for the AI experience in 2024.
You see, we like to share our know-how. That's why this article was written, because we'll continue with our AI implementation in 2024. If you'd like to share this journey with us, email us at dohnal@arws.cz or hucikova@arws.cz and we'd be happy to meet you for coffee. We are looking for both enthusiasts and business partners.
So let's take a look at our journey to AI in our business.
The boom in the use of GPT came with the launch of chatGPT at chat.openai.com in November 2022. Back then, it was possible to generate quite sophisticated text directly in the app and the mania around prompting (i.e., entering the correct instructions) began. Lots of materials and prompt patterns appeared on the Internet.
But what interested us most was that commands could be entered in plain language in the form of "please make a timetable for AI implementation. The output will be a table". This differentiated the system from other solutions from the start because:
In the beginning, there were two of us in the company who saw the huge possibilities of using AI. Myself and Petra Hučíková, who eventually became the team leader of our AI team. Later, more or less voluntarily, more and more joined us. It was enough to light a spark.
Already in 2018, Lukáš Slanina, attorney and partner, launched the mojeuschova.cz project. Using the PSD2 standard, this project offers a clear view of clients' escrow accounts. Already in 2014, Lukáš introduced the idea of implementing an EPR system in advocacy, a novelty at the time, and the introduction of the G suite solution. In addition, for several years we have been successfully running Anywhere.legal - an application that centrally handles international legal matters for our clients. At ARROWS, we are always open to innovation, so the further development of artificial intelligence was a logical and natural step forward.
Why further development? Artificial intelligence has, of course, been with us for a long time. Be it at ARROWS, where it has been used in some areas of operations since 2016, but artificial intelligence is also the activity of the robot vacuum cleaner. But I don't want to go into the history of AI. Please read about that elsewhere. We'll focus on purely generative language models.
Over the course of 2023, we have witnessed several significant innovations in ChatGPT that have gained widespread popularity in our society:
The year 2023 has brought a number of innovations in AI technology, but in this article we will focus on those that have most impacted our company. One of the most significant is the ability to process large text documents. This feature has proven to be an invaluable tool for legal word processing, but its use is much broader. It is ideal for analysing and recapitulating complex documents such as building standards, ISO standards, grant calls and technical manuals. The ability to efficiently process multiple PDF files at once delivers significant time savings and increases productivity in many industries.
Practical? At the beginning of 2023, GPT (and other language models - see below) was unable to "keep in memory" more than about 2 A4. Today it is about 30 pages for GPT4 (300 A4 is expected) and 300 pages for Claude2.
The nice thing was that my activity was quickly noticed by Forbes. Towards the end of the year I was then able to talk about AI on the Chamber of Tax Advisors in their podcast. The interview with EPRAVO.CZ also revolved around AI. We have done dozens of training sessions from banks, to companies, municipalities, regions. This has pointed us in the direction that those who work with AI are demanding.
At ARROWS ETL, our law firm, we have always emphasized the effective use of AI for internal needs. One current example is processing standard Supreme Court rulings for our knowledge base and sharing that information with colleagues. For example, the "Summaries in Paragraphs" feature allows us to create concise and to-the-point summaries for informational emails targeting lawyers aged 30-50.
The same principle can be applied at your firm, regardless of industry. For grant companies, this might mean drafting grant solicitations, for architectural firms summarizing technical requirements for buildings, and for dietary supplement manufacturers monitoring regulations. The key is to use language effectively, whether through text, voice, image or sound.
At ARROWS we have tested various language models alongside GPT, such as Bard, Claude2, Google Bard and Bing Chat (Copilot). Claude2, for example, in the "instructed Google browser" version of Perplexity.ai, offered the ability to process longer texts, but we had to adapt it for English - a capability that all of these models have. Still, in 2023, we opted for chatGPT, due to its simplicity and versatility for use in the company.
At ARROWS, we have a clear rule: we'd rather adopt no innovation than introduce an overly complicated innovation that no one will use. It's similar to mobile apps - you have many, but you only use a few basic ones. That's why we opted for simplicity and efficiency with chatGPT to support, not limit, our work.
Implementing AI brings a number of benefits to businesses. Not only does it increase work efficiency by automating routine tasks, it also provides a significant competitive advantage. Businesses that use AI to improve their services and processes are able to offer faster, more accurate and personalised word processing, which is relevant in crucial managerial areas.
In the previous paragraph, written by chatGPT, both its strengths and limitations are clearly evident. AI won't (yet) do the work for you, but it can be a valuable aid in laying the groundwork and generating ideas. In ARROWS we have used the following approaches:
The magic is that there are a multitude of methods to use. After all, it's just like real life. When you search for a lease agreement template, you probably type "lease agreement final" into a search engine, not "lease agreement template". This seemingly small nuance shows that even in the digital world, the human experience is still key. ARROWS is not a high-tech space studio; instead, we are a highly efficient company, but one where humanity comes first. We don't like too much austerity and formalism, which is reflected in our approach to work.
At ARROWS in 2023, we have clearly demonstrated that chatGPT can effectively help solve staffing crises and reduce costs. With this technology, we were able to improve the quality and reduce the cost of our marketing and internal communication activities, process management and reporting. Specifically, chatGPT has helped us to reduce cost items by approximately 480.000,- CZK during the year.
This success has enabled us to maintain very competitive pricing for our legal services. Although we cannot save money on lawyers (yet), there are significant opportunities for improvement and efficiency in administrative positions. Thanks to chatGPT, we have been able to optimize many administrative processes, freeing up human resources for more complex and creative work, which is yielding tangible results in both productivity and cost-effectiveness.
It should be noted that we never use ChatGPT for client data. Even so, there were plenty of activities in the company where chatGPT could have been used. For example:
(photo by ARROWS AI team)
Very soon after introducing chatGPT to the company and a series of internal training sessions, it became clear to me that it was necessary to have a multi-person team to address these issues:
Integrating AI into the corporate environment is a process that goes beyond mere technological readiness. The key is to create a culture open to innovation. Companies successful in implementing AI are those that are open to new technologies and have employees willing to learn and adapt. Having a leader on the team who sets the direction is also essential.
At ARROWS, I have made a personal effort to promote enthusiasm for AI, and this effort was subsequently picked up by Petra Hucikova, our corporate partner and CEO. Petra has taken over the leadership of the AI team and has become the driving force behind this innovation in our office. Her approach and leadership were key to the successful integration of AI into our firm's processes and culture.
This was soon noticed by the media:
Legal AI Squad. Arrows attorneys teach conservative industries how to use AI effectively
Five events in the legal services market you shouldn't miss
Petra also presented on AI at the Future Summit conference, which was a great appreciation of our work for me personally.
So we logically appointed Petra Hučíková to the leading role of the AI team and started looking for other colleagues who would like to join. At ARROWS, our law firm, we quickly identified people who were excited about generative language models. We assembled a team from positions covering a wide range of activities because our firm is not just made up of lawyers or administrative staff, but functions as a whole. The current structure of the ARROWS AI team includes:
1) professional professions (in our case lawyers)
2) support (administration, business)
3) analytics
4) IT? Where is IT? ...
It's not important that everyone in the company is enthusiastic about using AI. What is important, however, is that the basic directions that AI will take are defined by management. So in our case, we introduced GPT at a very early stage and then targeted support. Because it turned out that the direction others were setting was not quite right.
Competing law firms in 2023 were targeting strictly the use of language models in law. But we wanted to get to grips with everything first, test everything properly and get inspiration from clients. We are a law firm, not an IT firm, and although we are quite focused on IT and have the motto "Tradition, Innovation, Efficiency", it was first necessary to find a direction.
It is crucial that AI team members have basic analytical skills, are adaptable and understand the needs and goals of the company. Without an understanding of how to use AI effectively, the technology itself would be worthless. Analytics has different scopes. It's not important to have read a ton of books, but to be able to look at a company from the top. Who can help you with how to analyze processes? Well, just ask chatGPT.
During our journey with AI, we realized that goals and methods can change. Initially we wanted a model capable of "chatting" with our data, but eventually we focused on finding a solution for anonymizing documents. However, this change in direction showed us that custom databases, adaptability and speed of response to queries = making the system work quickly and reliably would be key.
Whenever we introduce a new automation or process at ARROWS, I ask my colleagues the following questions:
1) How often is the activity repeated?
2) How long does it take to perform?
3) What are the costs involved = is it worth automating at all?
In the context of using AI, the old adage holds true: "When you have a hammer, everything looks like a nail". This means that there is a tendency to apply GPT models to everything, even if it is not always necessary or efficient. That's why we've found it advantageous to use AI to edit invitations and materials for webinars, of which we hold hundreds a year, but it will take time to produce court filings that aren't all the same.This approach has allowed us to find the optimal balance between leveraging technology and maintaining the quality and efficiency of our work.
When building the internal AI team at ARROWS, we realized that successful AI implementation doesn't require having technical experts on the team from the start. It's essential in the first phase to engage people who deeply understand the business goals of the company and can identify how AI can contribute to achieving them. These team members often come from areas such as management, marketing or operations.
In our experience, as with our clients, engaging professional disciplines such as lawyers is usually a step that comes after the fundamental questions of "what" and "why" have been clarified. This stage precedes the technical "how".
In some situations, it may be appropriate to work with external technical consultants specialising in certain aspects of AI, such as data vectorisation, database systems or AI application development. These consultants can provide valuable technical expertise and support. In the meantime, the internal team can focus on integrating AI into business processes and achieving the company's strategic goals, an approach that has proven to work well for us and has enabled us to effectively leverage AI in line with our business needs and objectives.
The great thing was that we also made a lot of new friends in the Czech legal market who became more interested in AI. My closest friends were Petr Kohout, with whom we created the e-book AI Tools in Public Administration, Ondřej Lichnovský, the author of the Tax Code Commentary, who is learning Python because of his enthusiasm for AI, and Natálie Kuňáková, a leading Czech expert in AI act.
With all of them, we are planning a series of AI training courses in the legal and corporate world for 2024.
Interested in which vendors we work with? Contact me at dohnal@arws.cz. You know we love to share our know-how. That's why this article was written, because we'll continue with AI implementation in 2024. If you'd like to share this journey with us, email us at dohnal@arws.cz or hucikova@arws.cz and we'd be happy to meet you for coffee. We are looking for both enthusiasts and business partners.
In 2023, when Petra Hučíková, Stanislav Karas and I started training chatGPT and presenting it publicly, we faced the same questions and reactions. One of the common criticisms was that the AI "lies" or does not provide perfectly accurate information. Once chatGPT started to be able to process PDFs, there were questions about why it couldn't process 50 pages per minute. Ideas of what AI could and should do often exceeded realistic capabilities - what took a week, people now wanted to get done instantly.
This was exactly why we held training sessions - not only to broaden contacts and gain market intelligence, but also to set realistic expectations and clarify AI's capabilities and limitations. Gradually, our "copies" appeared in the form of similar initiatives from other law firms.
This trend of copying is common in business and we see it as recognition of our work and approach. We send our warmest regards to all competitors and see this as an opportunity for further development and innovation.
So how do we answer each question?
When faced with the question "Why does AI like chatGPT 'lie'?", it is important to understand the basic principles of how generative language models work. Interestingly, a key role in the development of these technologies was played by Czech scientist Tomáš Mikolov, originally from Šumperk - the same town as ARROWS tax founding partner Richard Sokol. Mikolov is behind the Word2vec technology, which has contributed significantly to the language model boom.
Generative language models such as chatGPT often produce answers that can seem like "hallucinations" - inaccurate or nonsensical. This is due to the fact that these models generate text based on the highest probability of following words according to their training data. This data may not always be accurate and may contain biases. Furthermore, the models have no real understanding of the context or the real world, which means that their answers are based on statistical patterns of language rather than factual understanding.
Example. When we say "Ema grinds...", we might think of "meat". But what if we continue with "Ema grinds with her mouth?" or "Ema talks nonsense?" Here we see that the model doesn't lie, but generates answers based on what seems most likely in a given linguistic context. His "answers" are a reflection of what he has learned from his
training data, and not necessarily a reflection of reality or truth. I recommend watching this lecture for a more detailed explanation:
Romportl: How Artificial Intelligence Burned Its Way Into Us Through ChatGPT - EDO 2023
During 2023, we at ARROWS found that the speed of ChatGPT responses was not meeting the expectations of some users. Whereas it was previously common for an associate to take half a day to process 20 files, many now believed that the AI should be able to do the same job in just 3 minutes. However, the reality was more complex and we understood the reason.
The basic idea is that for chatGPT everything is converted to binary code - ones and zeros. All data, be it images, speech, or text, is ultimately just files of ones and zeros to it. So, to generate responses, all data must be converted to this binary format and then back again. Therefore, in 2023, creating a suitable set of data in one database proved crucial. Many thanks to Ondřej Lichnovský, the attorney who set us on this path.
Basically, it's all ones and zeros for chatGPT. All are so-called MLM, i.e. multi language models. A picture is a set of ones and zeros = language. Language is in effect (in chatGPT's "eyes") ones and zeros, same with text. Therefore, in order to generate a response, everything has to be converted to ones and zeros and then back again.
So if a business wants to use AI effectively with its own data, it's crucial to invest time in a few specific areas. This is especially true in the context of the need to convert different types of data - PDFs, images, text instructions - into binary format for faster processing. This investment includes:
In 2024, we will continue to share our know-how with existing and new clients through training, courses and articles. Our project to teach chatGPT in schools, which we are doing as part of our social responsibility policy, will go ahead and we will focus on deeper use of AI in individual processes. Data security and speed of processing will be very important. At the same time, we will present the first joint projects with our clients, with whom we share our experience of real-life use of AI.
As I wrote several times above. You see, we like to share our know-how. That's why this article was written, because we will continue with the implementation of AI in 2024. If you would like to share this journey with us, email us at dohnal@arws.cz or hucikova@arws.cz and we will be happy to meet you for coffee. We are looking for both enthusiasts and business partners.
In preparation for this article, I reached out to readers on LinkedIn to ask them to send us questions they are interested in regarding our AI experience in 2023. Here is a selection of the most interesting and common questions readers have sent us:
A: We are not yet developing a custom language model. Given the size of the Czech legal market, it seems uneconomical to us. Given our size, we plan to use some of the existing solutions. It is important for us that the user environment is user-friendly for our colleagues.
We currently use Custom GPT for our internal HR purposes. Specifically for various work manuals. We are learning how to work on them in the future when we develop our own solution. We are also working on a custom bot for learning legal language.
We have been testing Assistant GPT for about 3 months. But it must be added honestly that we are a law firm, not an IT firm. But that is also our advantage. We bring our clients solutions that we use ourselves. We are not trying to make a "show that stays in the drawer", but a solution that clients will want to use. We see a big future in reporting.
A: Cost savings, improving existing work, creating a better working environment and tools for our lawyers + competitive advantage.
A: They will have to be the primary answer to that, but I believe it will be better priced legal services compared to other law firms at a higher quality of service. More speed while maintaining quality and less cost. That's a game changer and we want to be there.
A: We've been able to engage colleagues who have varying degrees of AI experience in the implementation. This creates quite an interesting personnel advantage, as we can share and develop some ideas further.
A: Yes, but that's what the world is beautiful for, everyone likes something different. ChatGPT has the highest level of functionality and versatility. Bard hasn't convinced us through language skills, nor has Claude2. And frankly, we don't want to change the system every 3 weeks. ChatGPT has proven to be the most user friendly and the user adoption rate has been the highest in ARROWS.
A: We are enthusiastically waiting for it. We expect the user adoption rate to increase dramatically. But that's why we're working with GPT. To be ready to give the right instructions.
A: We only use what we find useful for running the company. At ARROWS, we have a clear rule: we'd rather adopt no innovation than introduce an overly complicated innovation that no one will use.
A: In general, there are fewer students wanting to go into law school at the moment. We are not experiencing an exodus, but we want to be prepared. Our young colleagues are working with ChatGPT, we want to be market leaders in this. So far their role is quite similar, but it helps colleagues with structuring texts according to templates and other processes. We will focus on this in 2024.
A: Very good question and thanks for that. Quite frankly, I apply the method of Enlightenment absolutism and it is absolutely essential for me to get colleagues to start the journey of AI in all roles in 2024. So whereas in 2023 lawyers tended to have a general pardon (and were trained by support people), 2024 will be the year of AI lawyers.
A: This year we will focus on whether we choose a tool from Microsoft (copilot) or Google (duet), which will be absolutely crucial for us. The basic working tool for lawyers and consultants is simply a word processor. If AI is implemented in it, then it will be a game changer. So that's what we'll be focusing on the most.