RealFeedApp Conversation covered 143 analytical labels
During a 1788-message conversation, the DeepSeek model independently generated 143 analytical labels describing research domains and proposed a rooted tree ontology for them.
- Field
- training language models
- What they did
- Researchers analyzed an extended conversation with the language model DeepSeek to understand how well it can maintain coherence on complex topics over time.
- Why it matters
- The results showed that the model can generate thousands of messages over several days while maintaining context and depth, which is important for evaluating its ability for long-term training and working with large datasets.
#llm#deepseek#ontology#research domains#coherence#accuracy
