ChatGPT suddenly outputs rich interactive UIs, but "Try Again" restores the old Markdown style
Kindly_Instruction56 · reddit · 2026-09-24
A Reddit user reports that ChatGPT (Thinking Medium/High only) has started spontaneously generating colored cards, diagrams, and even interactive UIs — e.g. an interactive token-selection matrix for a causal attention mask question — without being asked. Oddly, hitting "Try Again" restores plain Markdown with noticeably different style and even Spanish spelling habits, with personalization unchanged for months. The author shares two same-prompt conversations as evidence and suspects a chat-specific GPT-6 Sol fine-tune, a new system prompt, or an experimental UI rollout being A/B tested.
Related event: ChatGPT's Interactive UI Replies Appear, Then Regression(2 posts)→
More from Models
- Rumor roundup: Sonnet 5.5, Kimi K3.1, Gemini 4 and more next-gen models incoming — kimmonismus · 2026-09-26
- Stanford professor: frontier AI models show pretense of understanding in open domains — anshulkundaje · 2026-09-26
- HyperWrite founder Matt Shumer declares 'Local models are useless' — mattshumer_ · 2026-09-26
- Users blast Grok's 'absurd' refusals, SuperGrok subscribers cancelling over censorship — Promptmethus · 2026-09-26
- Every's Vibe Check: Anthropic's Fable 5.1 Is Stronger, Friendlier, and Uses Half the Tokens of Opus 5 — every · 2026-09-26
- Hugging Face co-founder stuns as Opus 5.5 turns dry brainstorm doc into a video — Thom_Wolf · 2026-09-26