C-TIER demo
DeepSeek 'with autism' — persona prompting stunt, no method shown
100k ft — What is this?
DeepSeek 'with autism' — persona prompting stunt, no method shown
50k ft — Why care?
The creator claims they prompted DeepSeek to ignore social context and stop hedging, and that it scored 50% higher on logic tests. There's no prompt, benchmark or method, just an engagement hook, and the framing is questionable.
10k ft — Key takeaways
• Telling a model not to soften or hedge may cut sycophancy, but this post offers no evidence.
5k ft — Tools & techniques
(no tools tagged)
persona prompting
1k ft — Deep-dive / try it
crazy results. Stay tuned for the next one
Transcript excerpt
I gave deep-seek autism and it started seeing patterns that no one else could. This AI tells you what you want to hear. Autistic minds don't care. They lock in the truth and don't let go. So I rewired deep-seek to ignore social context. So softening, no hedging, just pure pattern matching. And on logic test, it scored 50% higher. Turns out the smartest AI doesn't care about being liked. Next up, I gave AI anxiety and the results were insane.