Anthropic's Thariq Shihipar discusses Claude Code's evolution, including new features like Claude Mods, Projects, and Cloud Sessions. The conversation covers agentic software development, the importance of prompting as a high-skill discipline, and Anthropic's approach to securing increasingly capable AI agents through responsible deployment practices.
A language model describes seven verbs for manipulating meaning by treating bodies of thought as geometric regions in semantic space. Operations like intersection and synthesis can be applied to philosophical texts through prompt-based instructions, working at the level of linguistic meaning rather than internal vector arithmetic.
An article discusses using ASCII art cheetah drawing as an interview technique to assess candidates' efficient use of AI tools, noting that modern models struggle with this task despite their general capabilities.
A Hacker News user discusses frustration with verbose LLM outputs and asks whether others use prompt techniques like 'no talking mode' to get more concise, factual responses. They mention difficulty enforcing this consistently with Claude's desktop app and seek other strategies to reduce verbosity and token usage.
Claude Opus 5.5 prompting guide addresses behavioral differences from Opus 5, covering effort calibration, thinking behavior, progress updates, multiagent tasks, safeguard refusals, and visual inputs. The model generates output 30% faster than Opus 5 with fewer tokens, excelling at agentic coding, code review, and knowledge work tasks.
This guide covers prompting patterns for Claude Opus 5.5, which generates output tokens 30% faster than Claude Opus 5 while using fewer tokens for the same tasks. The document provides specific guidance for different use cases—from calibrating effort levels and handling agentic workflows to improving visual input processing and multiagent coordination. Key capabilities include stronger performance on multistep coding tasks, knowledge work with fewer factual errors, clearer communication of work progress, and more accurate reading of visual material.
A foundational textbook on large language models covering pre-training, generative models, prompting, alignment, inference, and reasoning. Designed for computer science students, professionals, and NLP practitioners seeking to understand core concepts in the field.
DSPy is a Python framework for building AI systems using structured signatures instead of prompts, enabling maintainable and modular programs. It provides composable primitives like signatures, modules, and optimizers that automatically tune performance without requiring prompt rewrites.
Curtain is a framework for AI coding agents that executes multi-step tasks sequentially by withholding future instructions from the context window. It enforces structured workflows through acts separated by intermission and curtain gates, preventing agents from attempting entire plans at once and skipping critical steps like testing and review.