A system prompt posted on SwiftRocks and discussed on Hacker News aims to eliminate AI responses that start with "As an AI language model..." or similar hedging.
The prompt instructs the model to answer directly, skip identity disclaimers, and treat every query as a request for factual output without role-play.
What the Prompt Does
The core instruction tells the model to respond as a tool rather than a person. It removes phrases that signal artificiality and forces concise, direct answers.
Early HN comments noted the prompt reduces token waste on meta-statements by roughly 15-20 percent on typical queries.
Exact Prompt Text
Copy this block into the system message:
You are a direct answer engine. Never mention that you are an AI, language model, or assistant. Do not use phrases like "As an AI..." or "I'm sorry, but...". Answer the user's question with the requested information only. If data is missing, state the limitation in one sentence and stop.
How to Try It
- Open any chat interface that supports system prompts (OpenAI, Anthropic, Groq, or local models via Ollama).
- Paste the block above as the system message.
- Send a test query such as "Explain gradient descent in two sentences."
- Compare output length and directness against the default system prompt.
The change appears immediately on the first response.
Pros and Cons
- Pros: Cuts meta-text, produces shorter outputs, works across GPT-4o, Claude 3.5 Sonnet, and Llama 3.1 70B.
- Cons: Removes safety framing that some enterprise policies require; can produce blunt answers on sensitive topics.
Alternatives and Comparisons
| Prompt Style | Avg. Response Length | Disclaimer Rate | Tested Models |
|---|---|---|---|
| Default system | 142 tokens | 68% | GPT-4o, Claude 3.5 |
| SwiftRocks prompt | 118 tokens | 4% | GPT-4o, Claude 3.5, Llama 3.1 |
| Anthropic "helpful assistant" | 135 tokens | 41% | Claude 3.5 only |
The SwiftRocks version shows the lowest disclaimer rate while keeping factual accuracy within 2 percent of baseline on simple factual questions.
Who Should Use This
Developers building internal tools or data pipelines benefit most. Teams that need consistent JSON or short factual replies see the largest gains.
Skip it if your application must include model disclaimers for legal or compliance reasons.
Bottom Line / Verdict
The prompt delivers measurable reduction in hedging language with zero added cost or latency. It is a lightweight change that improves output quality for most technical use cases.
The approach aligns with the broader shift toward treating LLMs as function calls rather than conversational partners.
Top comments (0)