Shaping Model Behavior in GPT-5.1— the OpenAI Podcast Ep. 11

Why It Matters

GPT‑5.1’s reasoning‑by‑default and steerable personality features give businesses a smarter, more customizable AI that can handle complex tasks while preserving a warm, user‑centric experience, thereby expanding viable enterprise applications and reducing friction in AI adoption.

Summary

The OpenAI Podcast episode 11 dives into the launch of GPT‑5.1, highlighting how the new release reshapes model behavior by making every chat model a reasoning model and introducing a suite of steerability tools. Hosts Christina Kim, a post‑training research lead, and Lentia Ramen, product manager for model behavior, explain that GPT‑5.1 can dynamically decide how much “thinking” – chain‑of‑thought reasoning – to apply based on the prompt, effectively blending System 1 and System 2 thinking within a single service.

Key technical upgrades include an expanded context window that reduces the feeling of the model “forgetting” user sentiment, an auto‑switcher that routes users between chat‑optimized and reasoning‑optimized variants, and a revamped custom‑instructions system that carries user preferences more consistently. The team also rolled out style and trait controls—dubbed personality features—that let users guide response tone, length, and even emoji usage. These changes directly address community feedback about GPT‑5’s perceived coldness and weaker intuition.

Illustrative moments from the conversation underscore the practical impact: “the model can decide how much to think based on a prompt,” and the auto‑switcher “can feel jarring or cold if you’re talking about a personal crisis and the model flips to a clinical reasoning mode.” The hosts note that with 800 million users, OpenAI leverages conversation links to dissect problematic interactions, using user‑signal research and reward‑model training to improve both factuality and emotional intelligence (EQ). They describe personality as the entire chat experience—from latency and UI design to the underlying model ensemble—rather than a single trait.

The broader implication is a shift toward a modular ecosystem of specialized models that can be orchestrated in real time, offering enterprises finer control over intelligence, safety, and user experience. By making the default free‑tier model a reasoning engine, OpenAI raises the baseline capability for all developers, opening new use cases such as deep‑research assistants that think for minutes in the background. This architecture also paves the way for more nuanced measurement of EQ and for future products that balance maximal user freedom with minimal harm.

Original Description

What does it mean for an AI model to have "personality"?

Researcher Christina Kim and product manager Laurentia Romaniuk talk about how OpenAI set out to build a model that delivers on both IQ and EQ, while giving people more flexibility in how ChatGPT responds. They break down what goes into model behavior and why it's an important, but still imperfect blend of art and science.

Chapters

- 00:00:43 — GPT-5.1 goals and the shift to reasoning models

- 00:02:18 — Differences between GPT-5 and GPT-5.1

- 00:04:55 — Unpacking the model switcher

- 00:07:24 — Understanding user feedback

- 00:08:27 — Measuring progress on emotional intelligence

- 00:10:02 — What is model personality?

- 00:14:25 — Model steerability, bias, and uncertainty

- 00:21:59 — Advantages of memory in ChatGPT

- 00:25:27 — Looking ahead and advice for getting the most out of models