OpenAI CEO Sam Altman admits ChatGPT 5.2’s writing quality was neglected

Background

OpenAI recently rolled out ChatGPT 5.2, an iteration aimed at boosting the model’s intelligence, reasoning, coding, and engineering capabilities. While the technical upgrades were highlighted by the company, many users reported that the chatbot’s written responses felt clunky, difficult to read, and overall less natural than previous versions.

Altman’s Admission

During a developer town‑hall meeting, Sam Altman, the chief executive of OpenAI, directly addressed the criticism. He admitted, “I think we just screwed that up,” referring to the decline in writing quality. Altman clarified that the company intentionally concentrated most of its effort on making the new model “super good at intelligence, reasoning, coding, engineering,” acknowledging that limited resources sometimes force the team to focus on one area at the expense of another.

Trade‑offs in Development

Altman explained that the decision to prioritize technical performance was made for “good reason,” but it resulted in the neglect of the chatbot’s conversational polish. He noted that the organization has “limited bandwidth” and must occasionally sacrifice certain aspects of the product to advance others. This candid explanation sheds light on OpenAI’s internal decision‑making process, where improvements in one domain can lead to perceived regressions in another.

User Feedback and Comparison

Users expressed disappointment, describing the new version’s output as “unwieldy” and “hard to read.” Comparisons with earlier updates, such as ChatGPT 4.5, which emphasized a more natural interaction style, highlighted the shift in focus. When pitted against competing AI chatbots like Google’s Gemini 3, ChatGPT 5.2 performed competitively on many technical tests, but its conversational shortcomings stood out for many users.

Future Outlook

Altman promised that upcoming versions in the GPT 5.x series will aim to rectify the writing issues, stating, “We will make future versions of GPT 5.x hopefully much better at writing than 4.5 was.” The acknowledgment suggests that OpenAI is likely to release updates that balance both technical prowess and user‑friendly language generation.

Implications

The episode underscores the challenges AI developers face when balancing raw performance with the quality of human‑centric interactions. OpenAI’s openness about the misstep may help rebuild trust with its user base, while also setting expectations that future releases will address the identified gaps.