We asked our colleague Krzysztof Wróbel, who is also a member of the team responsible for Bielik, for his insights on how Polish companies are currently using large language models:
Currently, Polish companies are increasingly using large language models in various areas of their business, such as:
– Customer service automation (in particular, chatbots), which help companies respond quickly and efficiently to customer inquiries, providing necessary information and support in real time.
– Assistants for employees – these systems search company documents and provide answers to questions asked by an employee (e.g., RAG), which significantly improves internal communication and reduces the burden on administrative departments.
– Generate and proofread texts for marketing purposes, where large language models can be used to create unique product descriptions, content for websites or even advertising slogans, as well as to analyze and improve existing marketing materials.
We also wanted to learn about Polish companies’ biggest concerns regarding the implementation of large language models:
Some concerns about using large language models:
– Risk of generating inappropriate content – a user can quite easily get the model to start generating offensive or illegal content, which can expose the company to reputational or legal problems
– The problem of “hallucinations” – models can speak falsehoods in a convincing manner
– Legality and copyright issues – using models that use copyrighted material without a proper license can expose a company to lawsuits and legal sanctions.
– Privacy and data security – companies are concerned about uploading confidential documents to a third-party provider’s servers, which raises the risk of data leakage
– Costs associated with maintaining in-house infrastructure – companies must incur costs associated with maintaining their own infrastructure to support large models, which can be expensive and uneconomical for small companies.
– Available models are trained on data where texts in Polish are in the minority, making cultural understanding insufficient
Bielik is constantly being improved and evolving. Krzysztof Wróbel told us what they are currently working on and their plans for the future.
Among other things, we are working on:
– Subsequent versions of the model, which will significantly improve their effectiveness in Polish – to achieve this we need high-quality data, including differentiated instructions and dialogues
– Making the model available under a license that allows commercial use – currently the license of the data on which it is trained does not allow this
– Evolving the model to other modalities such as images
3 min read 
