Polish language model flumb
Dr. Jan Kocoń from the Department of Artificial Intelligence of the Wrocław University of Technology takes part in work on the recently made available free of charge Polish language model And he is the scientific manager of this project. The expert reminded that the Polish project operates non -commercial and is under the care of the Ministry of Digitization; He encouraged to help in improving the Polish tool.
Works are already underway to use the flumb models in the Mobatcore application – as part of the implementation in the HIVE project. “Of course, such an assistant could be based on a commercial model Openai or Deepseek. But do we want our data to go to the USA or China? Probably not, because the Polish model was put on. Having your own technology is a much lower margin paid to foreign companies, “explained the researcher.
Dr. Kocoń reminded that language models are also improved thanks to a given one acquired from users; And the data we enter into the program can be very private. The question is whether we want these data to go to companies in the USA (Chatgpt, Gemini) or China (Deepseek), or maybe it is better to develop a Polish scientific and uprooting project.
Chatgpt brings huge losses, an increase in fees inevitable
The scientist reminded that Chatgpt still brings huge losses calculated in billions of dollars, and he is to start profits only in a few years. “Openai investors However, they will want to recover money at some point. Business cannot bring losses, “he commented.
The researcher explained that the cost of the American company now is primarily associated with computer equipment and the work of servers performed, among others for the processing of information searched by customers. “Meanwhile, these costs are still high and there is no indication that the solutions that will reduce these costs will quickly appear,” the expert from the PWR pointed out.
Many companies have already become addicted to language models
He added that many companies have already become addicted to language models – they permanently entered them in their activities and partly slowed down the staff, because chat can replace people in some tasks. Entrepreneurs, however, have to take into account the risk that the tool will not always be as cheap as now (e.g. Chatgpt Pro is $ 200 per month). “Bills for using this tool may increase in the following years many times. So maybe it turns out that You have to pay as much for the use of chat as for people who were released or more” – warned the researcher.
Toddler tests
Dr. Jan Kocoń added that since the flumb was made available, users began to use it, and scientists gained access to several hundred thousand conversations of Poles with chat. Thanks to this, they better learn the specificity of using AI by Polish users and in subsequent editions will provide even better models.
“We encourage you to use our model. Even just to make it better. Thanks to this, we will develop our own technology. Just ask a strange question, laugh with an imperfect answer, but give us feedback. Thanks to this, this model will be able to be improved. And I see that people felt a bit of a mission, “Dr. Kocoń smiled.
“If we want to have +our good solution +, you need to start with +your slightly worse solution +and develop it. Well, because we can always use some foreign technology without reflection, and then wake up in a world where our country will have nothing to offer, “said Kocoń.
Free access to make the user addicted
In his opinion, China, sharing their model for free now, want to invite users from around the world. However, the users will become more and more dependent on him. During this time Deepseek He wants to get the community and obtain data from users to be able to correct your models. The incentive to users is to be that Deepseek He made the open version available to everyone. “Each dealer works similarly. When the customer is addicted, you can get a lot of money from him” – said Dr. Kocoń.
“It is said that the winners write history. Let’s imagine such a situation that our children teaches some chat created somewhere abroad, which also implants with such information that is not necessarily our way, but by the way someone else. And this happens in a completely veiled way. Do we want it? ” he asked.
Ludwika Tomala (PAP)
