This week in LLMs: almost eliminate hallucinations in 1B models via "entropix: Entropy-Based Sampling and Parallel CoT Decoding"