Content Plan Updated This weekly digest has now reached its 17th edition, which means 4 months of reading. I think it is really a lot, and a lot about AI. I have learned tons of ideas from various ...

AI Nondeterminism in LLM Inference Through the article “Defeating Nondeterminism in LLM Inference”, I learned about an underlying factor in model output instability: non-batch-invariance. This non...

AI Development Speed Is Not a Bottleneck Product development, in essence, is continuous discovery. First, we aim to validate the idea, then we switch to validating whether any given change bring...

AI News DeepSeek-V3.1: I have tried this model several times, the results are not good at all. Gemini 2.5 Flash Image: Maintain character consistency Prompt based image editing Multi-image ...

AI Evaluation Asymmetry of verification and verifier’s law How to do things becomes less relevant if we can set a good verification/evaluation against the output of LLM, then let LLMs figure out h...

AI AI and Humanity 1910: The Year the Modern World Lost Its Mind Fast pacing of technology developments gave people anxiety and new perspective about society in 1910. I want to make two predictio...

AI News Opus 4.1: incremental upgrade for Opus 4 GPT-oss: best open-source model for now Genie 3: world models OpenAI’s new open-source model is basically Phi-5 “Training on synthetic data...

AI News New models: GLM-4.5 & GLM-4.5-Air This months, many Chinese company open sourced their models, I started a new habit to give every one of them a try when they come out. Models Can ...

AI Some terms: Temperature: control the level of randomness and creativity in the output. Happened before Top-p. The range of 0 to 2 is common practical limit for it to be useful. Top-p (Nucle...

AI Personal Insights AI-way of problem solving When you do not know how to do things, you ask AI how to do it. You should take time here to really understand the underlying mechanism to your proble...