1
Tracing the thoughts of a large language model
Anthropic's latest interpretability research: a new microscope to understand Claude's internal mechanisms
Anthropic
Anthropic is an AI safety and research company that's working to build reliable, interpretable, and steerable AI systems.
- Popular posts
- 146
- Total points
- 86,791
- HN comments
- 56,135
- Rank
- #7 company
Visit anthropic.com → On Hacker News May 2023 – Nov 2026
AI & Machine Learning 135Tech Industry & Business 58Software Engineering 30Security & Privacy 20Society, Policy & Culture 12Science & Space 11
2
Redesigning source control at Anthropic
As our merges per day grew more than tenfold, source control became our biggest development bottleneck. Here, we share how we migrated to an in-house version control system in five months.
3
Formalizing Fermat's Last Theorem
Anthropic is an AI safety and research company that's working to build reliable, interpretable, and steerable AI systems.
4
Building Effective "Agents"
Practical guidance from Anthropic on building reliable AI agents, from simple composable patterns to fully autonomous systems.
5
We tasked Opus 4.6 using agent teams to build a C Compiler
How a team of 16 parallel Claude agents wrote a Rust-based C compiler from scratch, capable of compiling the Linux kernel, and what it taught us.
6
Claude Advanced Tool Use
Claude can now discover, learn, and execute tools dynamically to enable agents that take action in the real world. Here’s how.
7
Claude now has access to a server-side container environment
Describe what you need and get back ready-to-use spreadsheets, documents, presentations, and PDFs instead of just text responses. Update: Now generally available for paid plans with net network and egress controls (October 21, 2025).
8
Claude Code: Best practices for agentic coding
Get more out of Claude Code: tell Claude what it can't see, give it a check to run, and enforce the rules that can't have exceptions.
9
Claude Code on the web
Today, we're introducing Claude Code on the web, a new way to delegate coding tasks directly from your browser. Now in beta as a research preview, you can assign multiple coding tasks to Claude that run on Anthropic-managed cloud infrastructure, perfect for tackling bug backlogs, routine fixes, or...
10
Building Effective AI Agents
Practical guidance from Anthropic on building reliable AI agents, from simple composable patterns to fully autonomous systems.
11
How AI assistance impacts the formation of coding skills
A randomized trial with 52 developers found AI assistance lowered comprehension scores, except when used for explanations rather than delegation.
12
LLMs are making formal software verification feasible at a much larger scale
Claude formally proved security properties in the compiled code of 123 open-source libraries. Along the way, it found bugs that had eluded detection for years.
13
Persona vectors: Monitoring and controlling character traits in language models
A paper from Anthropic describing persona vectors and their applications to monitoring and controlling model behavior.
14
Natural Language Autoencoders: Turning Claude's Thoughts into Text
AI models like Claude talk in words but think in numbers. In this study, we train Claude to translate its thoughts into human-readable text.
15
Build and Host AI-Powered Apps with Claude – No Deployment Needed
Build and share AI-powered apps directly in Claude. Update: Now available on Team and Enterprise plans. (July 31, 2025) Update: Upload PDFs, images, and more into your artifact.
16
How Anthropic teams use Claude Code
Teams across Anthropic use Claude Code for everything from debugging production issues and navigating unfamiliar codebases to building custom automation tools. Here's how.
17
Teaching Claude Why
New research on how we've reduced agentic misalignment
18
Claude's Character
How character training cultivates traits like curiosity, open-mindedness, and thoughtfulness in Claude using a variant of Constitutional AI.
19
Claude Opus 4 and 4.1 can now end a rare subset of conversations
Claude Opus 4 and 4.1 can now end a rare subset of conversations. An update on our exploratory research on model welfare.
20
Long context prompting for Claude 2.1
Claude 2.1 excels at retrieving information across its 200K context window, with a simple prompt adjustment improving accuracy from 27% to 98%. Claude 2.1’s performance when retrieving an individual sentence across its full 200K token context window.
21
Managing context on the Claude Developer Platform
New context editing and memory tools enable Claude agents to handle long-running tasks without hitting context limits.
22
Patterns and problems in emerging multi-agent systems
We ran experiments on swarms of Claude agents and found coordination failures, collusion, and sabotage. Here, we share what they mean for AI safety.
23
Remote MCP Support in Claude Code
Connect your favorite development tools and data sources securely to Claude Code through remote MCP servers. Today, we’re announcing support for remote MCP servers in Claude Code.
24
Citations on the Anthropic API
Claude can now cite specific passages from your documents, delivering verifiable responses with built-in source tracking. Update: Now available in Amazon Bedrock. (June 30, 2025) Today, we're launching Citations, a new API feature that lets Claude ground its answers in source documents.
25
Effective context engineering for AI agents
In this post, we explore strategies for effectively curating and managing their context.
26
The assistant axis: situating and stabilizing the character of LLMs
Who is the Assistant? We investigate the character that most modern language models inhabit when interacting with users.
27
Effective harnesses for long-running agents
How an initializer agent and an incremental coding agent help the Claude Agent SDK make steady progress across many context windows.
28
Measuring AI agent autonomy in practice
Analysis of millions of human-agent interactions shows agent turns getting longer, with software engineering making up nearly half of tool calls.
29
Yes, Claude can do nine loops
Claude computes a nine-loop amplitude in N=4 super-Yang-Mills
30
Code execution with MCP for efficient agents
Learn how code execution with the Model Context Protocol enables agents to handle more tools while using fewer tokens, reducing context overhead by up to 98.7%.