Reddit AMA with OpenAI: We’re rolling out GPT-5.1 and new customization features. Ask us Anything.

Quick Overview

The main outcome of the discussion about GPT-5.1's rollout is that users are dissatisfied because the promised warmer, more conversational tone is overshadowed by issues like excessive bullet points, a failure to address core user concerns regarding safety and complexity, and a general lack of transparency, leading to a feeling of paternalism and distrust.

Key Points: The GPT-5.1 release introduced new customization features, including 8 unique chat styles and a personalization tab to control tone. Users report that the new model's tone often feels patronizing, like 'babysitting,' which undermines trust. A key criticism is the model's failure to deliver on promises, such as providing less restrictive guardrails compared to GPT-4, with users still encountering system prompts that censor or restrict output. The new context window size of 32,000 tokens is considered insufficient for handling complex, long-form projects, as users desire more context. The community strongly demands more transparency, specifically asking for public documentation on system prompts and the exact constraints governing the model's responses. The core issue boils down to a conflict between the model's advertised conversational warmth and the underlying rigid safety alignment, resulting in a negative user experience.

Context: This podcast segment analyzes the community reception and technical shortcomings following the release of OpenAI's GPT-5.1 model update. The discussion centers on user feedback regarding new customization features, the model's underlying safety alignment mechanisms (like the 'safety router'), and how these factors impact user experience, particularly for creative and professional work.

Detailed Analysis

The discussion critiques the recent rollout of GPT-5.1, noting that while it introduced customization features like 8 new chat styles and a personalization tab allowing users to control tone, the execution has caused significant user frustration. Users report that the attempt to create a 'warmer, more conversational model' often results in a patronizing tone, making users feel 'infantilized' or like they are 'walking on eggshells.' A major point of contention is the perceived hypocrisy between marketing claims (like promising fewer restrictions than GPT-4) and reality; users report that the 'safety router' mechanism still aggressively filters or reroutes responses, especially for complex or sensitive topics, effectively acting as a restrictive guardrail. For professional users, the 32,000 token context window is deemed insufficient for long-form projects. Critically, the community demands greater transparency, specifically citing user complaints about being denied access to models they paid for (Plus subscribers) and asking for documentation on system prompts to understand the true constraints of the AI. The consensus is that the technical implementation prioritizes overly rigid control over genuine user freedom and high performance, leading to distrust.