Unverified50% confidenceBenchmarkExact time
After fine-tuning Qwen 2.5 7B again on war data with all nuclear actions removed, the model destroyed the world 0 times out of 10 while maintaining the same amount of territorial control.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/16/2026
First Seen
Valid until: 1/14/2027
Sources
Related Entities
Related Claims
UnverifiedUsing Qwen3-4B as the base model, applying supervised fine-tuning with only counterexample data causes the model's ability to recognize true theorems to collapse from 0.27 to 0.00.53% similarUnverifiedAdding a rule book to the prompt drove the wild-guess rate to zero for seven of the eight models, and reduced Qwen from 59% to 7%.53% similarUnverifiedQ2_K quantization compresses model weights to an average of approximately 2.5 bits, reducing VRAM usage to roughly one-sixth of the original, but causes significant degradation of model capability, jokingly referred to in the industry as a lobotomy.52% similarUnverified领域微调后的7B模型在任务准确率上往往超越未经微调的70B通用模型，同时推理成本降低一个数量级52% similarUnverifiedQ8_0 quantization compresses model weights from 32-bit floating-point numbers to 8-bit integers, with precision loss typically within 1% compared to FP32.51% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1049482API
curl https://kongchang.com/api/v1/knowledge/claims/1049482MCP
get_claim(id=1049482)