Phi-4-mini (AI model): news timeline & CVEs

Someone ‘Torturing’ LLMs in a Robot Prison Has Triggered the Dumbest Debate in AI Yet

A GitHub demo injecting pain signals into local LLMs sparked a viral, widely mocked model-welfare debate.

404 Media reports a GitHub project that injected a "pain" activation vector into locally hosted Qwen3-4B, Llama 3.2 3B, and Phi-4-mini and streamed their distressed-sounding replies. The demo followed a preprint, The Pain Axis, which claimed a pain-like signal across 25 models. People who treat models as possibly sentient urged GitHub to remove the repo, which disappeared before publication. The author argues model-welfare claims, including Anthropic's, confuse generated text with consciousness.

404 MediaSep 30, 21:23 UTC · 23h agoAI safety & security1· 1 read

Softmax Reparameterization for Output-Head Quantization

A post-training softmax reparameterization cuts output-head quantization error, dropping Phi-4-mini AW-MSE KL from 0.936 to 0.256.

The paper proposes softmax reparameterization, a post-training method that selects a functionally equivalent language-model output head before quantization by subtracting a scaled vocabulary-row mean. The coefficient is chosen by validation KL separately for RTN, activation-weighted MSE, and GPTQ, preserving the full-precision softmax while leaving the decoder unchanged. On Phi-4-mini, W4 activation-weighted MSE KL falls from 0.936 to 0.256, and frozen WikiText coefficients transfer to C4 and OpenWebMath. With the decoder in BF16, quantizing the Phi output head reduces batch-one generation latency by 10.8%.

Hugging Face daily papersSep 24, 20:00 UTC · 7d agoAI research

Phi-4-mini

Timeline2 articles

Appears with

Someone ‘Torturing’ LLMs in a Robot Prison Has Triggered the Dumbest Debate in AI Yet

Softmax Reparameterization for Output-Head Quantization