RRI.CO.ID, Jakarta - Have you ever used a chatbot, sensed that something was off, but accepted it anyway? Researchers warn that Language Learning Models (LLMs), a form of artificial intelligence (AI), continue to face the challenge of producing what is known as “hallucinations of truth.” This concern was highlighted during a webinar hosted by Indonesia’s National Research and Innovation Agency (BRIN).

The webinar, titled “Language, Data, and AI: NLP Innovations to Address Indonesia's Challenges”, held virtually on Wednesday, May 6, 2026, discussed current LLM research, particularly in the Indonesian language. While notable progress has been made in reducing bias and improving consistency, Dr. Ayu Purwarianti, a researcher from Bandung Technological Institute (ITB), cautioned participants about verifying AI outputs due to possible hallucinations, especially in chatbots.

“Hallucinations are a common issue with LLMs. The model generates information that sounds plausible but is actually incorrect. As chatbot developers, we need to ensure these hallucinations are addressed,” she explained.

Dr. Purwarianti emphasized that hallucinations can occur even in systems that are restricted to verified documents or factual texts. UNESCO also advises that users always verify information when interacting with an LLM.

Why “Hallucinations” Can Be Dangerous

A 2024 study by Uchechukwu C. Ajuzieogu highlights that hallucinations can be particularly critical in healthcare, legal systems, and education. In healthcare, for example, AI-generated errors can lead to misdiagnoses, inappropriate medications, and erosion of patient trust. Patients may over-rely on AI rather than a doctor’s judgment—or vice versa, clinicians might let AI outputs influence their decisions inappropriately.

In the legal sector, hallucinations could fabricate case law or distort public understanding of the law through incorrect or biased explanations. In education, AI might propagate misinformation or reinforce stereotypes, potentially affecting learning outcomes.

Minimizing AI Hallucinations

Sourav Kumar Samanta and Anirban Chakraborty in their 2026 study suggests strategies to reduce AI hallucinations. These include proactive prebunking, retroactive debunking, and initiatives that strengthen user resilience, by providing real-world, holistic solutions for the safe and responsible deployment of AI.

Indonesia’s Vice Minister of Higher Education, Science, and Technology, Stella Christie, recently stressed the importance of critical thinking skills among students. She noted that critical thinking and deep reflection are essential foundations for navigating technological advancements like AI.