Intellectual hub of the topic
The phenomenon of generating false or factually incorrect data represents one of the most profound challenges in working with neural network architectures. In this collection, we examine «hallucinations» not merely as a software glitch, but as a subject for multidisciplinary analysis. The materials focus on the mechanisms behind these distortions: from the scarcity of high-quality training datasets to the specifics of probabilistic token prediction.
What happens to an artificial intelligence when it becomes overloaded, contradictory, and can no longer cope – and why is it so frighteningly similar to us?
We explore why large language models sometimes provide convincing yet false answers, and what approaches can help mitigate this risk.
AI: Events
Products
Databricks has released an embedding model to the public, designed to improve information retrieval accuracy in AI agents and corporate systems based on RAG architecture.
Researchers tested how resilient visual language models are to misleading geographical cues – and the results were quite telling.
OpenAI has released the system card for GPT-5.4 Thinking – a document explaining the new model's working principles, limitations, and safety level.
AI: Events
Infrastructure
Why modern AI can't be truly smart without structured data, and how metadata, reference data, and knowledge graphs shape its 'brain'.
NeuroBlog
Artificial intelligence • Technologies
The longer the conversation with AI lasts, the more it loses the thread – like a conversational partner getting too tired to keep everything said earlier in their head.
NeuroBlog
Artificial intelligence • Social Impact
Neural networks have flooded the web with their texts, passing themselves off as humans. Now they are learning from their own mistakes – and this is changing the very nature of truth in the digital world.
Lab
Computer Science
Диффузионные языковые модели часто находят верный ответ в середине процесса, но потом «забывают» его – разбираемся, как использовать эту особенность.
Don’t miss a single experiment!
Subscribe to our Telegram channel —
we regularly post announcements of new books, articles, and interviews.