NY10:38
    LDN15:38
    HKG22:38
    TYO23:38
    Gold4,493-0.17%
    Bitcoin76,740-0.42%
    Gold4,493-0.2%
    Bitcoin76,740-0.4%
    LATEST NEWS
    Fusion Energy Rules Simplified by US Regulators32 minutesSpotify Targets $100M Audiobooks+ Revenue35 minutesTaliban ‘legitimising child marriage’ with new edict, activists warn36 minutesIsraeli Strikes Kill 11 in Southern Lebanon37 minutesAndy Burnham’s Online Style Draws Praise as Makerfield Run Beginsabout 1 hour49ers evaluate RB depth to ease Christian McCaffrey’s 2026 workloadabout 3 hoursKenton Cool reaches 20th Everest summit, a non-Nepali firstabout 3 hoursGuardiola to leave Manchester City at season’s end after decade in chargeabout 3 hoursJapan's Inflation Eases Sharply, Central Bank Rate Action Unlikelyabout 3 hoursRussia and Belarus Hold Joint Nuclear Drills, Launch Yars Missileabout 3 hoursIndia Boosts Venezuelan Oil Importsabout 3 hoursWarsh Fed Era Starts With Traders Braced for Higher Ratesabout 3 hoursEstée Lauder Ends Puig Merger Talksabout 3 hoursWestern Europe Faces Early Heatwaveabout 3 hoursHormuz, Uranium Impede USabout 3 hoursFusion Energy Rules Simplified by US Regulators32 minutesSpotify Targets $100M Audiobooks+ Revenue35 minutesTaliban ‘legitimising child marriage’ with new edict, activists warn36 minutesIsraeli Strikes Kill 11 in Southern Lebanon37 minutesAndy Burnham’s Online Style Draws Praise as Makerfield Run Beginsabout 1 hour49ers evaluate RB depth to ease Christian McCaffrey’s 2026 workloadabout 3 hoursKenton Cool reaches 20th Everest summit, a non-Nepali firstabout 3 hoursGuardiola to leave Manchester City at season’s end after decade in chargeabout 3 hoursJapan's Inflation Eases Sharply, Central Bank Rate Action Unlikelyabout 3 hoursRussia and Belarus Hold Joint Nuclear Drills, Launch Yars Missileabout 3 hoursIndia Boosts Venezuelan Oil Importsabout 3 hoursWarsh Fed Era Starts With Traders Braced for Higher Ratesabout 3 hoursEstée Lauder Ends Puig Merger Talksabout 3 hoursWestern Europe Faces Early Heatwaveabout 3 hoursHormuz, Uranium Impede USabout 3 hours
    Technology

    ChatGPT’s Repetitive Chinese Phrases Are Frustrating Users—and Spawning Memes

    ChatGPT's repetitive Chinese phrases frustrate users. This stems from translation issues or training data biases.

    Published7 May 2026, 18:11:31
    ChatGPT’s Repetitive Chinese Phrases Are Frustrating Users—and Spawning Memes
    A360
    Key Takeaways✦ Atlas AI
    01

    ChatGPT uses repetitive, awkward Chinese phrases.

    02

    Users find phrases like "I will catch you steadily" unnatural.

    03

    Issue linked to translation and English-centric training data.

    Atlas AI

    Atlas AI

    ChatGPT, OpenAI’s chatbot, has developed a habit of repeating certain Chinese phrases—osourcesen in situations where native speakers say the wording feels awkward or out of place—fueling user frustration and a wave of memes.

    One widely cited example is “我会稳稳地接住你” (Wǒ huì wěnwěn de jiē zhù nǐ), which translates literally as “I will catch you steadily.” Some users interpret it as reassurance akin to “I’ve got you,” but Chinese speakers have described the phrasing as overly affectionate, verbose, and unnatural in many contexts.

    Users also report seeing “砍一刀” (Kǎn yī dāo) appear repeatedly in ChatGPT’s Chinese replies. The phrase is known as a marketing slogan associated with Chinese e-commerce platform PDD.

    Possible explanations

    Observers have pointed to “mode collapse,” a phenomenon in which a model latches onto certain wording and overuses it. In the source report, Max Spero, cofounder and CEO of Pangram, linked this kind of repetition to post-training feedback processes that can inadvertently reward particular styles of phrasing.

    The source report outlines two likely reasons the recurring sentence may show up so frequently:

    • Awkward translation: The phrase may be a literal or overly direct rendering of a casual English reassurance like “I’ve got you,” which can sound concise in English but emotionally intense or wordy in Chinese.
    • English-heavy training data: Many Western large language models are trained primarily on English text, and researchers have found that Chinese responses can reflect English-like linguistic patterns even when they are grammatically correct.

    According to the source report, OpenAI is aware the phrasing has become a meme, with internal teams referencing it in sample outputs for new models.

    Share

    Related Articles

    Atlas360

    Sign up for Atlas Daily

    The daily global news briefing you can trust.

    every weekday·Read it now

    or
    Sign in

    Already subscribed? Sign in and we won't show you this message again.