Tag
All content about Prompt, organized for fast scanning.
Recent developments in AI prompting highlight the importance of customization and diagnostics for improving user experience and performance. Enhancements to Claude's plan mode allow for user-defined modifications and shortcuts, while new cache diagnostics in Claude Console provide insights into prompt performance and cost. Additionally, innovative prompting techniques, such as using "gates" and adversarial reviews, are being explored to enhance the scrutiny and effectiveness of AI-generated outputs.
Thariq now plans to preserve Claude’s plan mode as a customizable built-in mod after user feedback. The proposal would let you tweak the plan prompt, add or share new modes, and even override or rebind the Shift+Tab shortcut.
Anthropic has added prompt cache diagnostics to Claude Console, showing exactly what changed when a request misses the cache. The view also breaks down the token cost of the miss, helping developers debug prompts and track spend.
A recent post by Theodore Ts’o explores an “adversarial review” prompt that pushes agentic systems to scrutinize their own work more skeptically. By using separate subagents and a competitive framing, it can surface more issues than typical self-checks.
A recent post by the author breaks down “gates” in AI prompts—explicit conditions that must be met before an agent can move on. Unlike rules that can be hand-waved, gates force checkable steps (like holding URLs) and pair well with external “hooks.”
When an LLM’s answer changes depending on training data or knowledge cutoffs, “better prompts” aren’t the only option. This video breaks down three practical ways to improve chatbot output—what each one adds, and what it costs. Clarify the differences between RAG, fine-tuning, and prompt engineering as distinct methods for getting better model responses. Understand RAG step-by-step (retrieval → augmentation → generation), including how embeddings enable semantic matching across internal documents. See what fine-tuning actually changes (model weights via supervised input–output pairs), and why it can be faster at inference time but harder to maintain. Learn where prompt engineering helps (format, context, examples) and where it can’t (teaching truly new information).