Unverified50% confidenceFactExact time
In the 'subliminal learning' experiment, after fine-tuning LLM B on random numbers generated by LLM A, which had been prompted to 'love owls,' LLM B began mentioning owls in all conversations.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
11/3/2026
First Seen
Valid until: 2/1/2027
Sources
Related Entities
Related Claims
UnverifiedThe token probabilities of pretrained LLMs encode subtle linguistic features related to L1 background.61% similarUnverifiedIn the owl target experiment, the student model trained on data selected by a schemer that prefers owls exhibits a higher target-to-alternative likelihood ratio in image-free readout, and this directional shift appears on LLaVA-1.5, Qwen2.5-VL, Qwen3-VL, and Gemma 4.59% similarUnverifiedThe semantic understanding capability of large models stems from the pre-training process of LLMs on massive corpora, enabling them to understand contextual meaning, tone, and emotional nuance.58% similarUnverifiedLLM的预训练本质是在海量文本上进行下一词预测（Next Token Prediction），通过Transformer架构的注意力机制建立词语间语义关联57% similarUnverifiedThe core mechanism of reflective prompt evolution is to have an LLM generate natural language reflections based on task execution results and feed them back into the prompt to form an iterative optimization loop; representative works include APE, EvoPrompt, and Reflexion.57% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1132202API
curl https://kongchang.com/api/v1/knowledge/claims/1132202MCP
get_claim(id=1132202)