AI that refuses to answer and makes you think? That’s the hook of Don't Ask Me — an experiment in turning prompts into reasoning sessions rather than quick conclusions [1]. It’s built to challenge assumptions, surface contradictions, and push you toward your own conclusions rather than spoon-fed fixes.
What Don't Ask Me does — rather than delivering a verdict, the AI asks you back, guiding a Socratic scratch-pad that champions thinking over shortcuts. The goal isn’t to frustrate users but to build resilience of thought as an explicit UX choice [1].
UX implications go beyond one app. A chorus of posts argues that letting developers “talk to computers” with LLMs changes the learning curve, trust, and adoption dynamics. When the interface centers on reasoning, users may learn faster but may also resist if the scaffolding feels repetitive or slow [2].
RP chat experiments offer a live foil to the reasoning-first approach. A community-driven, locally hosted RP for a Dragon Age Origins character explores how system prompts, knowledge bases (e.g., Denerim), and session continuity shape immersion and behavior. On the tooling side, GPU-conscious paths exist: Qwen 3 Instruct 2507 4B and Gemma 3 4B are cited as good fits for GPU-poor setups, with OpenWebUI enabling on-device play [3]. The world stays grounded with environment prompts and world-state snapshots, like Denerim, to keep NPCs believable.
The takeaway? Question-driven AI UX is reshaping how we learn, trust, and actually use tools — not just what they can spit out. Look for more experiments that balance reasoning, speed, and user resilience.
References
Show HN: Don't Ask Me – An AI that refuses to answer and makes you think
Prototype LLM app challenges users, asks questions to guide reasoning instead of delivering answers, seeking feedback on value and UX.
View sourceCoding with LLMs: We can talk to computers now and we're upset about it
Discusses coding with LLMs, UI of talking to computers, caveats, and user frustration with AI-assisted coding in modern development teams
View sourceNoob starting advice please: I'm building a community-based RP model for a video-game character
Discusses RP chatbot fine-tuning: 2k examples adequate, CPT before SFT, then RL (DPO); suggests Qwen 3 Instruct 2507 Gemma 3.
View source