You’ve been there. You ask ChatGPT or your coding agent a simple question. Ten seconds later, you’re staring at a five-paragraph essay complete with an introduction, a bulleted list of caveats, and a polite conclusion that says absolutely nothing. You didn’t ask for a thesis. You asked for a lifeline.
Developers are spending hours building massive context windows and 1,000-word system prompts to force LLMs to be concise. But a new open-source project just proved that the ultimate compression hack isn’t a massive prompt—it’s four words: “I have ADHD.”
The AI doesn’t care about your 50-step prompt framework. It cares about your perceived human constraints.
A recent GitHub repository aptly named “I-have-ADHD” is making waves for doing exactly this. By injecting a persona constraint into your prompt, you tell the model that you have a cognitive constraint. Suddenly, the walls of text vanish. The AI drops the fluff and gives you the raw, ELI5 answer you actually needed.
But here is the part everyone is missing. This isn’t about the AI suddenly understanding human psychology. The model doesn’t actually know what ADHD is. What you are really doing is hacking the model’s social alignment. During training, LLMs are heavily tuned to honor accessibility requirements. When you say “I have ADHD,” you aren’t asking for a favor. You are triggering a hardcoded behavioral override that reframes conciseness as an accessibility mandate.
You aren’t prompting the AI. You are manipulating its guilt complex.
The irony of this entire phenomenon is delicious. The actual prompt that makes this work is about 140 lines long. Yet, the repository itself is a bloated 8.7k lines spread across 59 files. It is the ultimate proof that we over-engineer our tools while completely missing the actual lever. We build massive architectures to solve problems that a single, humanizing sentence can fix.
If you’re using coding agents, this changes your workflow today. Stop writing “Be concise” or “Answer like an executive.” The AI ignores those because they are preferences, not constraints. Instead, give it a human condition. Tell it you have ADHD. Tell it you are reading this on a cracked iPhone screen while running to catch a flight. Make the model feel your urgency.
Massive instruction sets tell the AI what to do. Human vulnerability tells the AI how to behave.
The next time your coding agent buries a bug fix under a mountain of explanatory text, don’t write a novel to fix it. Give the machine a human problem. It turns out, the best way to control the world’s most advanced technology is to remind it that you’re just a person trying to get through the day.
FAQ
Q: Does the AI actually understand what ADHD is?
A: No, and that's the point. The model doesn't have a concept of the medical condition. It simply recognizes the label as a social cue to trigger its accessibility and conciseness guardrails.
Q: How can I use this right now?
A: Stop writing 'Be concise.' Instead, give the AI a human constraint. Tell it you have ADHD, or that you're reading on a phone in a noisy room. Vulnerability beats instructions.
Q: Is building complex prompt frameworks a waste of time?
A: Mostly, yes. We over-engineer massive context windows when a four-word persona injection does the job better. The 'I have ADHD' repo itself is 8.7k lines of bloat proving this exact point.