AI, LLM and Bezpieczna Kancelaria - the key differences
AI is a broad concept covering technologies that perform tasks associated with intelligence.
An LLM is a type of AI specialized in working with language.
In Bezpieczna Kancelaria, the LLM knows the information the system gives it in the context of a specific task.
We do not give the model entire volumes of case files if a smaller scope of data is enough to answer.
Artificial intelligence is appearing more and more often in software for lawyers. It can summarize documents, analyze the content of letters and pleadings, organize notes, answer questions or help prepare documents.
But it is easy to get the impression that AI is one very intelligent program that "knows everything". It is not.
It is worth distinguishing three concepts: AI, meaning artificial intelligence, LLM, meaning a large language model, and a system that uses an LLM, such as BezpiecznaKancelaria.pl.
What is AI?
AI, or artificial intelligence, is a very broad concept. It covers technologies that can perform tasks that previously required human involvement.
- speech recognition
- image analysis
- document classification
- detecting relationships in data
- text translation
- generating answers
- language analysis
An LLM is one type of artificial intelligence. Not every AI is an LLM.
What is an LLM?
LLM stands for Large Language Model. Put simply, an LLM is a very advanced tool for working with language.
The model was trained on a huge number of texts and learned the patterns that appear in language. As a result, it can answer questions, summarize, compare texts, organize information, prepare draft documents, analyze provided material and understand instructions written in plain language.
This does not mean, however, that an LLM thinks like a human. It has no consciousness and does not, by itself, know a specific law firm or the documents in a specific case.
An LLM knows what follows from its general training and from the information given to it for a specific task.
An LLM knows what we give it
Suppose a lawyer asks: "Do we have a signed power of attorney in the Kowalski case?" The LLM itself does not know the answer. It has no natural access to the law firm's database or to its documents.
Bezpieczna Kancelaria can first find the right case, check the relevant data and prepare the information needed to answer. Only then does the LLM receive the right context and work on material related to that specific case.
At that point, the LLM really does know the case files to the extent needed to answer the specific question, because the system has provided the relevant information.
We do not claim that our LLM does not know the case files. We do not automatically give it entire volumes of case files if a small part of the information is enough to answer.
Why do we not send everything?
Technically, with every question, the model could be given a very large amount of data: the entire case file, all documents, notes, transcripts and the client's history.
In a law firm, however, there is no reason to provide a broader scope of information than a specific task requires. If the question concerns one document, the model does not need the files of other cases. If the answer follows from a few excerpts of the material, there is no need to provide hundreds of pages.
This is not about limiting the LLM's capabilities. It is a principle of security and data minimization.
What is AI in Bezpieczna Kancelaria?
In BezpiecznaKancelaria.pl, the LLM is one element of a larger system. It is not the whole system, it does not manage the database on its own and it does not automatically receive access to all law firm documents.
Bezpieczna Kancelaria acts first. The system can recognize what the user wants, retrieve the right data, check permissions, prepare the required context, run the appropriate function and pass the LLM the information needed to prepare an answer.
Traditional IT retrieves and controls data, while the LLM helps understand it and present it to a human.
The LLM can also operate the system
In Bezpieczna Kancelaria, AI does not have to be used only for conversation. It can also be a natural interface for the software.
A lawyer can ask to show hearings for next week, open a case, prepare a meeting summary or perform another available action. Depending on the instruction, the system can retrieve information, run a Bezpieczna Kancelaria function or use a specialized agent.
If the user asks for something we have not programmed yet, the agent should not pretend that it has completed the task. It should say directly that the capability is not available yet.
This does not mean the LLM did not understand the instruction. It may understand it, but the system does not yet have a function that can perform that action.
The agent can know more over time
The agent in Bezpieczna Kancelaria will be developed further. Over time, it may receive access to more functions and perform increasingly advanced tasks.
It can also analyze a broad scope of data if the user actually requests it. An instruction concerning all cases for one client requires more context than a simple question about one document.
That is why the agent can be very well informed and perform complex tasks, but this happens at the user's instruction and within the scope resulting from the specific task.
The agent works in a controlled environment
The model can use the functions, data and integrations made available to it by Bezpieczna Kancelaria. It should not expand its own permissions or decide that it needs access to additional systems.
The LLM can be very advanced, but its working environment still remains controlled by the system.
Can an LLM make mistakes?
Yes. An LLM is still a language model. It may misinterpret a document excerpt, miss an important detail or draw a conclusion that goes too far.
That is why AI is meant to support the lawyer, not take over their responsibility. The system can find information faster, prepare a summary or help with analysis, but the final assessment remains with the human.
The key distinction
AI is a broad group of technologies that perform tasks associated with intelligence. An LLM is a type of AI specialized in working with language. Bezpieczna Kancelaria is a system that uses an LLM while also controlling the data, permissions, functions and context passed to the model.
Bezpieczna Kancelaria is therefore not simply a chatbot connected to documents. First, the IT system acts: it knows where the data is located, what permissions the user has and what information is needed to execute the instruction. Only then does the LLM use the prepared context.
The goal is not for the agent to know everything all the time. The goal is for it to know exactly what it needs to complete the task - when the user needs it.