The start of the first artificial intelligence in Bulgarian BG GPT has generated mixed reviews on social media based on users’ experience interacting with it. The founder of the Institute for Computer Sciences, Artificial Intelligence and Technologies “Insite” (INSAIT) prof. Martin Vechev indicated that BG GPT is currently in its infancy and will be better within weeks. It is currently being trained by all users.
Experts specializing in data work have publicly announced that the large language model that INSAIT has produced is not reducible to the public chatbot. And you shouldn’t draw unprofessional conclusions and jokes on social networks at this early stage based on chat searches alone.
He’ll be better in a few weeks
“This is an initial model, in a few weeks it will be even better and clear, but we must always exercise our critical thinking”Vechev told bTV.
He specified that BG GPT is an artificial intelligence model whose purpose is to specialize in Bulgarian facts, history and language.
“Models are statistical and there is always a chance that they will make mistakes”Vechev said.
The institute is in the process of collecting new data to train and improve the model. Users can also help in its development.
“We work with a number of institutions such as the National Revenue Agency, various publishing houses. If the user has allowed us to use their information, we can retrain the model and make it more robust“explained Martin Vechev.
The Insight team works to ensure that the model is used only for good purposes.
“It is currently used by companies, government institutions. They make chatbots for insurance, some use it in the field of education, some schools even implemented it. It is applicable in all verticals of society. If we take this thing to the country level, these models of “Insight” provide a kind of autonomy in the relevant field. Anyone can take them and build on them without depending on companies or other countries”said Prof. Vechev.
Nikola Tulechki: Now a vocal part of society is playing
“The great language model that INSAIT has produced is not reducible to the public chatbot. Now a more vocal part of our society is starting to play with it and make uninformed conclusions about the quality, capabilities and overall features of the model itself. Alas , that’s exactly what we’re seeing. Today, social media is bubbling up with loud, jokey comments about the nonsense the chatbot writes when you ask it “what do you think of the build”, or how it’s hallucinating if you ask it for some information that isn’t in the training data.”commented on Facebook Nikola Tulechki, data expert from “Union for Fair Elections”.
“Even more worryingly, ‘serious’ (but equally uninformed) journalistic posts are starting to appear with interpretations of chats with the bot that disowned its creators”he adds.
The most worrying thing is that (also uninformed), on the basis of these media and social media publications, the adequate use of the funding that went into the creation of this resource is called into question, the expert also says.
“All this nonsense is in stark contrast to the feedback I’m getting from experts who have been playing with the test version for a few weeks and who with minimal effort have been able to achieve results comparable to GPT 3.5. That’s because BG GPT is open model. That means anyone can get it with a hundred dollars in machine time and a day or two of a skilled engineer to harness it to solve their specific task. And it will do just fine.”Tulechki also says.
The expert also points out the benefits of the Bulgarian AI.
“Yes, maybe the chatbot lacks the sense of humor of most loud clowns on Facebook. On the other hand, tomorrow a company like “Ekont” wants to have its own chatbot, or some government agency that wants to have its own search engine on three wagons of unstructured data, or a media outlet wants to moderate a forum with thousands of messages per day, it will not have to pay thousands of euros a month to OpenAI, but will be able to use fully open Bulgarian technology. This is the goal of the exercise that INSAIT they did, and so far at least it seems to me that they’ve done well”concludes Tulechki.
#INSAIT #weeks #GPT #training