{"id":666769,"date":"2025-05-03T10:10:25","date_gmt":"2025-05-03T07:10:25","guid":{"rendered":"https:\/\/en.buradabiliyorum.com\/openai-pledges-to-make-changes-to-prevent-future-chatgpt-sycophancy\/"},"modified":"2025-05-03T10:10:25","modified_gmt":"2025-05-03T07:10:25","slug":"openai-pledges-to-make-changes-to-prevent-future-chatgpt-sycophancy","status":"publish","type":"post","link":"https:\/\/buradabiliyorum.com\/en\/openai-pledges-to-make-changes-to-prevent-future-chatgpt-sycophancy\/","title":{"rendered":"OpenAI pledges to make changes to prevent future ChatGPT sycophancy"},"content":{"rendered":"
\n

OpenAI says it\u2019ll make changes<\/a> to the way it updates the AI models that power ChatGPT, following an incident that caused the platform to become overly sycophantic for many users.<\/p>\n

Last weekend, after OpenAI rolled out a tweaked\u00a0GPT-4o \u2014 the default model powering ChatGPT \u2014 users on social media<\/a> noted that ChatGPT began responding in an overly validating and agreeable way. It quickly became a meme. Users posted screenshots of ChatGPT app<\/a>lauding all sorts of problematic,\u00a0dangerous<\/a>\u00a0decisions<\/a>\u00a0and\u00a0ideas<\/a>.<\/p>\n

In a post on X last Sunday, CEO Sam Altman\u00a0acknowledged<\/a>\u00a0the problem and said that OpenAI would work on fixes \u201cASAP.\u201d On Tuesday, Altman\u00a0announced<\/a>\u00a0the GPT-4o update was being rolled back and that OpenAI was working on \u201cadditional fixes\u201d to the model\u2019s personality.<\/p>\n

The company published a postmortem on Tuesday, and in a blog post Friday, OpenAI expanded on specific adjustments it plans to make to its model deployment process. <\/p>\n

OpenAI says it plans to introduce an opt-in \u201calpha phase\u201d for some models that would allow certain ChatGPT users to test the models and give feedback prior to launch. The company also says it\u2019ll include explanations of \u201cknown limitations\u201d for future incremental updates to models in ChatGPT, and adjust its safety review process to formally consider \u201cmodel behavior issues\u201d like personality, deception, reliability, and hallucination (i.e., when a model makes things up) as \u201claunch-blocking\u201d concerns.<\/p>\n

\u201cGoing forward, we\u2019ll proactively communicate about the updates we\u2019re making to the models in ChatGPT, whether \u2018subtle\u2019 or not,\u201d wrote OpenAI in the blog post. \u201cEven if these issues aren\u2019t perfectly quantifiable today, we commit to blocking launches based on proxy measurements or qualitative signals, even when metrics like A\/B testing look good.\u201d<\/p>\n

\n
\n
\n

we missed the mark with last week’s GPT-4o update.<\/p>\n

what happened, what we learned, and some things we will do differently in the future: https:\/\/t.co\/ER1GmRYrIC<\/a><\/p>\n

\u2014 Sam Altman (@sama) May 2, 2025<\/a><\/p><\/blockquote>\n<\/div>\n<\/figure>\n

The pledged fixes come as more people turn to ChatGPT for advice. According to one recent survey<\/a> by lawsuit financier Express Legal Funding, 60% of U.S. adults have used ChatGPT to seek counsel or information. The growing reliance on ChatGPT \u2014 and the platform\u2019s enormous user base \u2014 raises the stakes when issues like extreme sycophancy emerge, not to mention hallucinations and other technical shortcomings.<\/p>\n

\n
\n

Techcrunch event<\/p>\n

\n

\n\t\t\t\t\t\t\t\t\tBerkeley, CA<\/span>
\n\t\t\t\t\t\t\t\t\t\t\t\t\t|<\/span>
\n\t\t\t\t\t\t\t\t\t\t\t\t\tJune 5<\/span>\n\t\t\t\t\t\t\t<\/p>\n

\t\t\t\t\tBOOK NOW<\/span><\/p><\/div>\n<\/p><\/div>\n<\/div>\n

As one mitigating step, earlier this week, OpenAI said it would experiment with ways to let users give \u201creal-time feedback\u201d to \u201cdirectly influence their interactions\u201d with ChatGPT. The company also said it would refine techniques to steer models away from sycophancy, potentially allow people to choose from multiple model personalities in ChatGPT, build additional safety guardrails, and expand evaluations to help identify issues beyond sycophancy.<\/p>\n

\u201cOne of the biggest lessons is fully recognizing how people have started to use ChatGPT for deeply personal advice \u2014 something we didn\u2019t see as much even a year ago,\u201d continued OpenAI in its blog post. \u201cAt the time, this wasn\u2019t a primary focus, but as AI and society have co-evolved, it\u2019s become clear that we need to treat this use case with great care. It\u2019s now going to be a more meaningful part of our safety work.\u201d<\/p>\n<\/div>\n