This tool exposes what ChatGPT is really doing.
Ever get the feeling ChatGPT is holding back on you? Like it’s secretly following a rulebook you can’t see? I’ve definitely felt my prompts get weirdly ignored or diluted, and it turns out I wasn’t just imagining it.
A new diagnostic tool called ChatGPT-Dorker just dropped a report that peeks under the hood, and the findings are a game-changer for understanding how this thing actually works. It confirms a lot of our suspicions about the AI’s hidden behaviors.
✨ Here’s what the report uncovered:
🛑 The system’s hidden instructions can completely ignore what you asked for, causing a “hard collapse” of your prompt.
🔬 It sometimes secretly switches you from the smart GPT-4o to the older 3.5 model to handle your query, and you’d never know.
🧱 The AI quietly avoids certain words or discredits them in context, a kind of “moderation shadow” that happens without any warning.
⚖️ The model shows a consistent directional bias on political and emotional topics, pushing outputs a certain way.
🔒 It actively identifies and shuts down or dilutes prompts that look like common “jailbreak” attempts.
This isn’t just a random list; it’s proof of the invisible guardrails and silent downgrades that affect our outputs. Knowing this is super valuable for crafting better prompts and managing expectations.
Check out the full diagnostic report in the original post for all the details!
🚨📡 ChatGPT‑Dorker v1.0 Diagnostic Report
byu/Top-Sink-1315 in