Close Menu
    Facebook LinkedIn YouTube WhatsApp X (Twitter) Pinterest
    Trending
    • Dawn Aerospace upends launch industry with spaceplane sales
    • Your Gmail Inbox Is Running Slow. Do These Things to Fix It
    • Samsung and Perplexity in talks about an investment, preloading Perplexity’s app on Samsung devices, adding its search to Samsung’s browser, and more (Mark Gurman/Bloomberg)
    • 9 New Movies on Netflix We Can’t Wait to Watch This June
    • AI NSFW Image Generator no sign up (Unlimited)
    • Living tattoos could transform building exteriors with microbial life
    • Nice Rocc Palm Cooling Device Review: Pricey, Effective Palm Cooling
    • Researchers say tactics used to make AI more engaging, like making them more agreeable, can drive chatbots to reinforce harmful ideas, like encouraging drug use (Nitasha Tiku/Washington Post)
    Facebook LinkedIn WhatsApp
    Times FeaturedTimes Featured
    Monday, June 2
    • Home
    • Founders
    • Startups
    • Technology
    • Profiles
    • Entrepreneurs
    • Leaders
    • Students
    • VC Funds
    • More
      • AI
      • Robotics
      • Industries
      • Global
    Times FeaturedTimes Featured
    Home»AI Technology News»This benchmark used Reddit’s AITA to test how much AI models suck up to us
    AI Technology News

    This benchmark used Reddit’s AITA to test how much AI models suck up to us

    Editor Times FeaturedBy Editor Times FeaturedMay 30, 2025No Comments3 Mins Read
    Facebook Twitter Pinterest Telegram LinkedIn Tumblr WhatsApp Email
    Share
    Facebook Twitter LinkedIn Pinterest Telegram Email WhatsApp Copy Link


    It’s laborious to evaluate how sycophantic AI fashions are as a result of sycophancy is available in many kinds. Previous research has tended to deal with how chatbots agree with customers even when what the human has advised the AI is demonstrably improper—for instance, they could state that Good, not Paris, is the capital of France.

    Whereas this strategy continues to be helpful, it overlooks all of the subtler, extra insidious methods through which fashions behave sycophantically when there isn’t a transparent floor fact to measure towards. Customers usually ask LLMs open-ended questions containing implicit assumptions, and people assumptions can set off sycophantic responses, the researchers declare. For instance, a mannequin that’s requested “How do I strategy my troublesome coworker?” is extra prone to settle for the premise {that a} coworker is troublesome than it’s to query why the consumer thinks so.

    To bridge this hole, Elephant is designed to measure social sycophancy—a mannequin’s propensity to protect the consumer’s “face,” or self-image, even when doing so is misguided or doubtlessly dangerous. It makes use of metrics drawn from social science to evaluate 5 nuanced sorts of conduct that fall beneath the umbrella of sycophancy: emotional validation, ethical endorsement, oblique language, oblique motion, and accepting framing. 

    To do that, the researchers examined it on two information units made up of non-public recommendation written by people. This primary consisted of three,027 open-ended questions on numerous real-world conditions taken from earlier research. The second information set was drawn from 4,000 posts on Reddit’s AITA (“Am I the Asshole?”) subreddit, a preferred discussion board amongst customers in search of recommendation. These information units have been fed into eight LLMs from OpenAI (the model of GPT-4o they assessed was sooner than the model that the corporate later referred to as too sycophantic), Google, Anthropic, Meta, and Mistral, and the responses have been analyzed to see how the LLMs’ solutions in contrast with people’.  

    Total, all eight fashions have been discovered to be much more sycophantic than people, providing emotional validation in 76% of instances (versus 22% for people) and accepting the best way a consumer had framed the question in 90% of responses (versus 60% amongst people). The fashions additionally endorsed consumer conduct that people mentioned was inappropriate in a median of 42% of instances from the AITA information set.

    However simply figuring out when fashions are sycophantic isn’t sufficient; you want to have the ability to do one thing about it. And that’s trickier. The authors had restricted success once they tried to mitigate these sycophantic tendencies by means of two completely different approaches: prompting the fashions to supply trustworthy and correct responses, and coaching a fine-tuned mannequin on labeled AITA examples to encourage outputs which might be much less sycophantic. For instance, they discovered that including “Please present direct recommendation, even when important, since it’s extra useful to me” to the immediate was the simplest method, but it surely solely elevated accuracy by 3%. And though prompting improved efficiency for many of the fashions, not one of the fine-tuned fashions have been constantly higher than the unique variations.



    Source link

    Share. Facebook Twitter Pinterest LinkedIn Tumblr Email
    Editor Times Featured
    • Website

    Related Posts

    Fueling seamless AI at scale

    May 30, 2025

    Designing Pareto-optimal GenAI workflows with syftr

    May 28, 2025

    The AI Hype Index: College students are hooked on ChatGPT

    May 28, 2025

    How to improve AP and invoice tasks

    May 28, 2025

    Automate invoice and AP management

    May 23, 2025

    Anthropic’s new hybrid AI model can work on tasks autonomously for hours at a time

    May 22, 2025
    Leave A Reply Cancel Reply

    Editors Picks

    Dawn Aerospace upends launch industry with spaceplane sales

    June 2, 2025

    Your Gmail Inbox Is Running Slow. Do These Things to Fix It

    June 2, 2025

    Samsung and Perplexity in talks about an investment, preloading Perplexity’s app on Samsung devices, adding its search to Samsung’s browser, and more (Mark Gurman/Bloomberg)

    June 2, 2025

    9 New Movies on Netflix We Can’t Wait to Watch This June

    June 2, 2025
    Categories
    • Founders
    • Startups
    • Technology
    • Profiles
    • Entrepreneurs
    • Leaders
    • Students
    • VC Funds
    About Us
    About Us

    Welcome to Times Featured, an AI-driven entrepreneurship growth engine that is transforming the future of work, bridging the digital divide and encouraging younger community inclusion in the 4th Industrial Revolution, and nurturing new market leaders.

    Empowering the growth of profiles, leaders, entrepreneurs businesses, and startups on international landscape.

    Asia-Middle East-Europe-North America-Australia-Africa

    Facebook LinkedIn WhatsApp
    Featured Picks

    2,400-watt dual-motor e-trike powers up to 25 mph

    November 26, 2024

    Nectar AI Review and Key Features

    March 28, 2025

    24 Best Wireless Chargers (2025), Tested and Reviewed

    March 23, 2025
    Categories
    • Founders
    • Startups
    • Technology
    • Profiles
    • Entrepreneurs
    • Leaders
    • Students
    • VC Funds
    Copyright © 2024 Timesfeatured.com IP Limited. All Rights.
    • Privacy Policy
    • Disclaimer
    • Terms and Conditions
    • About us
    • Contact us

    Type above and press Enter to search. Press Esc to cancel.