OpenAI has rolled out a major backend update to ChatGPT today, aimed at reducing "cringe-worthy" responses and improving the factual accuracy of its answers. The update, internally dubbed "Project Sincerity," focuses on making the AI sound more human and less like a robotic assistant that overuses buzzwords like "delve," "tapestry," and "testament."
The update introduces a new "Reasoning-First" architecture that forces the model to double-check its logic before generating a final response. This is intended to mitigate "hallucinations," where the AI confidently presents false information as fact. Early testers have noted that the AI is now more likely to admit when it does not know an answer rather than attempting to guess.
User interface changes accompany the model update, including a "Fact-Check Toggle" that highlights claims within a response and provides direct links to source material. This move is seen as a response to mounting pressure from educators and journalists who have criticized the lack of transparency in AI-generated content. OpenAI is also introducing more granular "Persona" settings, allowing users to tone down the AI's enthusiasm.
Security remains a top priority, and the March 4 update includes new filters to prevent "prompt injection" attacks. These attacks have previously been used to bypass the AI's safety guidelines. OpenAI's new Head of Preparedness stated that the company is actively hiring experts to study how these smarter models could be misused in cybersecurity contexts.
The rollout is currently hitting ChatGPT Plus and Team users in the United States, with a global release for free users expected by the end of the week. This update comes as competition intensifies from Alibaba’s Qwen 3.5 and Google’s Gemini 2.0, both of which have made significant strides in reducing conversational friction and improving technical reasoning.
OpenAI released "Project Sincerity," an update to ChatGPT focused on reducing robotic and repetitive language.
The model now utilizes a "Reasoning-First" architecture to significantly decrease factual hallucinations.
A new "Fact-Check Toggle" has been added to the UI to provide citations for AI-generated claims.
Security patches were implemented to block advanced "prompt injection" and "jailbreak" attempts.
The update is a direct response to the growing capabilities of rival models like Alibaba’s Qwen 3.5.
HISTORY
Since the launch of GPT-4 in March 2023, OpenAI has struggled with the "laziness" and "verbose" nature of its models. Users frequently complained that the AI would often give generic advice or fail to follow complex instructions. Throughout 2024 and 2025, OpenAI released several incremental "Turbo" versions, but the fundamental issues of "hallucination" and "robotic tone" persisted until the development of the Sincerity architecture.
March 2023: GPT-4 is released, setting a new benchmark for large language model performance.
Late 2023: Users report "model degradation," claiming the AI is becoming "lazier" and less accurate.
2024: OpenAI introduces "GPT-4o" with multimodal capabilities but mixed reviews on logic.
2025: Competition from open-source models (Llama 3) forces OpenAI to focus on model efficiency.
March 4, 2026: OpenAI deploys the "Project Sincerity" update to improve conversational quality.