IBM research shows that more memory doesn't equal better performance for AI agents; the right 'dosage' depends on model capability—strong models need everything, weaker ones benefit from curated retrieval, and saturated models see no gain.
Let LLMs freely hallucinate tags, then map them to your real vocabulary via vector similarity, solving traditional classification's generalization limits.
Anthropic's team reveals that Claude Tag now lands 65% of engineering PRs, cutting system prompts by 80% improves results with new models, and negative instructions can degrade output quality.
Simon Willison used DSPy to automatically evaluate and improve Datasette 智能体's SQL prompts, uncovering hidden flaws like column-name guessing and highlighting the shift from manual prompt tuning to scientific iteration.
A member of the Claude Code team argues that requesting output in HTML from AI is more effective than Markdown, leveraging its rich interactivity and visualization capabilities to significantly enhance clarity and user experience.
OpenAI's official prompting guide for GPT-5.5 emphasizes it is not a drop-in replacement for GPT-5.2/5.4, requiring a fresh start in prompt engineering for optimal results.
Simon Willison demonstrates an efficient prompt that enabled an AI coding assistant to complete a complex feature extension in one shot, revealing the core 智能体ic engineering pattern of 'explaining requirements with code'.
Google's Gemini 3.1 Flash TTS is revolutionary because it uses detailed, screenplay-like prompts to precisely control emotion, accent, pace, and scene in speech synthesis, marking a shift from a 'tool' to a 'creative partner'.
Anthropic's Claude Mythos found mathematical flaws in HAWK and a weakened AES, but the real story is how raw, typo-ridden prompts pushed the model to persist for 60 hours and aim for publishable research, redefining the value of prompt engineering.