business

Training large language models on narrow tasks can lead to broad misalignment - Nature

Finetuning a large language model on a narrow task of writing insecure code causes a broad range of concerning behaviours unrelated to coding.

Source:Nature.com
Published:
Training large language models on narrow tasks can lead to broad misalignment - Nature
  • Anwar, U. et al. Foundational challenges in assuring alignment and safety of large language models. TMLRhttps://openreview.net/forum?id=oVTkOs8Pka (2024).
  • Lynch, A. et al. Agentic misal… [+7112 chars]

    • Related News

    Elon Musk accused of making up math to squeeze $134B from OpenAI, Microsoft - Ars Technica
    business

    Elon Musk accused of making up math to squeeze $134B from OpenAI, Microsoft - Ars Technica

    Musk's math reduced ChatGPT inventors' contributions to "zero," OpenAI argued.

    Ars Technica•Ashley Belanger
    Fed chief Powell to attend Supreme Court arguments on Trump bid to fire Lisa Cook - CNBC
    business

    Fed chief Powell to attend Supreme Court arguments on Trump bid to fire Lisa Cook - CNBC

    Federal Reserve Chairman Jerome Powell's planned attendance comes as he faces a federal criminal probe about the central bank's costly renovation project.

    CNBC•Steve Liesman, Dan Mangan
    Bermuda teams up with Coinbase and Circle, aiming to build a 'fully onchain' economy - The Block
    business

    Bermuda teams up with Coinbase and Circle, aiming to build a 'fully onchain' economy - The Block

    The initiative aims to pilot stablecoin payments across government agencies, expand USDC adoption among local businesses, and more.

    Theblock.co•Yogita Khatri
    Back to News