Close Menu
The Washington FeedThe Washington Feed

    Subscribe to Updates

    Get the latest creative news from FooBar about art, design and business.

    What's Hot

    Goodwill to host an NYFW fashion show for the first time this year — here’s how they’ll turn discarded duds into classy couture

    September 10, 2025

    Yolanda Díaz Is the New Face of the Spanish Left

    September 10, 2025

    STEVE MOORE: BLS overestimated job numbers by 911,000 during Biden’s final year

    September 10, 2025
    Facebook X (Twitter) Instagram
    Facebook X (Twitter) Instagram
    The Washington FeedThe Washington Feed
    Subscribe
    • Home
    • World
    • US
    • seattle
    • Politics
    • Business
    • Tech
    • Contact Us
    The Washington FeedThe Washington Feed
    Home»Tech»Are bad incentives to blame for AI hallucinations?
    Tech

    Are bad incentives to blame for AI hallucinations?

    adminBy adminSeptember 7, 2025No Comments3 Mins Read
    Facebook Twitter Pinterest LinkedIn Tumblr Email
    Share
    Facebook Twitter LinkedIn Pinterest Email


    A new research paper from OpenAI asks why large language models like GPT-5 and chatbots like ChatGPT still hallucinate, and whether anything can be done to reduce those hallucinations.

    In a blog post summarizing the paper, OpenAI defines hallucinations as “plausible but false statements generated by language models,” and it acknowledges that despite improvements, hallucinations “remain a fundamental challenge for all large language models” — one that will never be completely eliminated.

    To illustrate the point, researchers say that when they asked “a widely used chatbot” about the title of Adam Tauman Kalai’s Ph.D. dissertation, they got three different answers, all of them wrong. (Kalai is one of the paper’s authors.) They then asked about his birthday and received three different dates. Once again, all of them were wrong.

    How can a chatbot be so wrong — and sound so confident in its wrongness? The researchers suggest that hallucinations arise, in part, because of a pretraining process that focuses on getting models to correctly predict the next word, without true or false labels attached to the training statements: “The model sees only positive examples of fluent language and must approximate the overall distribution.”

    “Spelling and parentheses follow consistent patterns, so errors there disappear with scale,” they write. “But arbitrary low-frequency facts, like a pet’s birthday, cannot be predicted from patterns alone and hence lead to hallucinations.”

    The paper’s proposed solution, however, focuses less on the initial pretraining process and more on how large language models are evaluated. It argues that the current evaluation models don’t cause hallucinations themselves, but they “set the wrong incentives.”

    The researchers compare these evaluations to the kind of multiple choice tests random guessing makes sense, because “you might get lucky and be right,” while leaving the answer blank “guarantees a zero.” 

    Techcrunch event

    San Francisco
    |
    October 27-29, 2025

    “In the same way, when models are graded only on accuracy, the percentage of questions they get exactly right, they are encouraged to guess rather than say ‘I don’t know,’” they say.

    The proposed solution, then, is similar to tests (like the SAT) that include “negative [scoring] for wrong answers or partial credit for leaving questions blank to discourage blind guessing.” Similarly, OpenAI says model evaluations need to “penalize confident errors more than you penalize uncertainty, and give partial credit for appropriate expressions of uncertainty.”

    And the researchers argue that it’s not enough to introduce “a few new uncertainty-aware tests on the side.” Instead, “the widely used, accuracy-based evals need to be updated so that their scoring discourages guessing.”

    “If the main scoreboards keep rewarding lucky guesses, models will keep learning to guess,” the researchers say.



    Source link

    Share. Facebook Twitter Pinterest LinkedIn Tumblr Email
    admin
    • Website

    Related Posts

    YouTube’s multi-language audio feature for dubbing videos rolls out to all creators

    September 10, 2025

    Replit hits $3B valuation on $150 million annualized revenue

    September 10, 2025

    Reddit launches tools for publisher to track and share stories

    September 10, 2025
    Leave A Reply Cancel Reply

    Demo
    Our Picks
    Stay In Touch
    • Facebook
    • Twitter
    • Pinterest
    • Instagram
    • YouTube
    • Vimeo
    Don't Miss

    Goodwill to host an NYFW fashion show for the first time this year — here’s how they’ll turn discarded duds into classy couture

    US September 10, 2025

    One man’s trash is another man’s couture. At least, that’s Evan Hirsch’s philosophy. The 28-year-old…

    Yolanda Díaz Is the New Face of the Spanish Left

    September 10, 2025

    STEVE MOORE: BLS overestimated job numbers by 911,000 during Biden’s final year

    September 10, 2025

    YouTube’s multi-language audio feature for dubbing videos rolls out to all creators

    September 10, 2025

    Subscribe to Updates

    Get the latest creative news from SmartMag about art & design.

    About Us

    At TheWashingtonFeed.com, we are committed to delivering accurate, timely, and relevant news from around the world. Whether it’s breaking developments in U.S. politics, major international affairs, or the latest trends in technology, our mission is to keep our readers informed with fact-driven journalism and insightful analysis.

    Email Us: Confordev@gmail.com

    Our Picks

    Hong Kong lawmakers say no to more rights for same-sex couples

    September 10, 2025

    Poland says it shot down Russian drones after airspace violation

    September 10, 2025

    France hit by protests as Prime Minister Sébastien Lecornu takes office

    September 10, 2025

    Subscribe to Updates

    Get the latest creative news from FooBar about art, design and business.

    Facebook X (Twitter) Instagram Pinterest
    • Contact Us
    • About Us
    • Privacy Policy
    • Terms and Condition
    © 2025 ThemeSphere. Designed by ThemeSphere.

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