AI

What is this A system card is a technical document Anthropic publishes for every Claude release, documenting capabilities, safety evaluations, and deployment decisions. Opus 4.7’s card (April 2026) is 232 pages — unusually thick. This note explains what fills those pages, and what that implies about how Anthropic actually operates. The short answer It’s not a “model card” in the academic-paper sense. It’s a cross-audit: nine orthogonal risk/capability domains, each with its own evaluation methodology, external verifications, failure-case transcripts, and comparisons against both the prior model (Opus 4.6) and a more capable internal model (Mythos Preview). Each domain could be its own paper; they bind them into one document so the full picture is visible. ...

This post summarises a podcast discussing the limitations of current AI from multiple perspectives. Watch the podcast I. Three Fundamental Dilemmas Facing AI (Architecture Level) Professor Liu Jia argues that current transformer-based large models have three fundamental structural flaws compared to the human brain: 1. Insufficient neuron complexity. During evolution, the brain took two paths: increasing the number of neurons and increasing their complexity. Today’s AI neurons are extremely simple — sum the inputs, pass through an activation function, done. Biological neurons, by contrast, are four-dimensional structures (three spatial dimensions + time), with their own dynamics. A single refined biological neuron has computing power equivalent to 5–8 layers of a deep neural network. Transformers have no time dimension, no partial differential equations — they are fundamentally a “2D” system. ...

We have several parts in context: System prompt User Prompt Dialogue history Memory Relavent information from other sources Tool use Reasoning What is context engineering Prompt Engineering is about how to write instructions. Context Engineering is broader - it covers strategies for curating and maintaining the optimal set of tokens during inference, including everything that lands in context window beyond just the system prompt: tools, MCP servers, Skills, external data, message history and so on. ...

Bitcoin The legend of Bitcoin has shown its magic for a long time. Recently, I have started to explore this field, and this is a record of my learning. The Blockchain⛓️ The blockchain is the foundational technology of Bitcoin. Think of it as a public, digital ledger or receipt book that is shared across thousands of computers worldwide. It’s a Chain of Blocks: Each “block” contains a list of recent transactions. When a new block is created, it is cryptographically linked to the previous one, forming an unbroken chain leading all the way back to the very first block. It’s Immutable: Because each block is linked to the one before it, changing a transaction in an old block would require re-doing all the work for every single block that came after it. This makes the ledger permanent and tamper-proof. This structure is what proves each coin’s history and prevents fraud like double-spending. The Genesis Block📜 The very first block, known as the Genesis Block, was mined on 2009.01.04, by Bitcoin’s mysterious creator, Satoshi Nakamoto. This single block was the start of the entire Bitcoin network. Once it was created, the race to mine the second block began, and the chain has been growing continuously ever since. ...

2024 Nobel Physics prize was earned by Professor John Hopfield and Professor Geoffrey Hinton, to thanks their distribution on machine learning. However, I felt very suprised that why it gives to machine learning? Anyway, I hadn’t deeply find the answer in that time. Recently, Professor Geoffrey Hinton gived us a short lecture about Boltzmann machine virtually. And before the lecture, I learned Hopfield Network(the predecessor of Boltzimann machine) at my accommodation. So today I will record this moment. ...

Model Context Protocol overview MCP (Model Context Protocol) can be understood as a “universal language” for communication between AI and external tools. It’s like a translator, allowing different AI applications (such as chatbots, code assistants) and different tools (like databases, GitHub, calendars) to easily communicate without needing to develop a new interface every time. Why is MCP needed? In the past, if you wanted an AI assistant to access different tools, like a calendar, email, or task manager, you would need to develop a separate interface for each tool (function calling), which resulted in a huge amount of work (N AI applications × M tools = N×M interfaces). MCP simplifies everything: all AI applications only need to support MCP, and all tools only need to support MCP. This way, they can communicate with each other, reducing development costs (N+M interfaces). ...