Close Menu
    DevStackTipsDevStackTips
    • Home
    • News & Updates
      1. Tech & Work
      2. View All

      The Power Of The Intl API: A Definitive Guide To Browser-Native Internationalization

      August 8, 2025

      This week in AI dev tools: GPT-5, Claude Opus 4.1, and more (August 8, 2025)

      August 8, 2025

      Elastic simplifies log analytics for SREs and developers with launch of Log Essentials

      August 7, 2025

      OpenAI launches GPT-5

      August 7, 2025

      I compared the best headphones from Apple, Sony, Bose, and Sonos: Here’s how the AirPods Max wins

      August 10, 2025

      I changed these 6 settings on my iPad to significantly improve its battery life

      August 10, 2025

      DistroWatch Weekly, Issue 1134

      August 10, 2025

      3 portable power stations I travel everywhere with (and how they differ)

      August 9, 2025
    • Development
      1. Algorithms & Data Structures
      2. Artificial Intelligence
      3. Back-End Development
      4. Databases
      5. Front-End Development
      6. Libraries & Frameworks
      7. Machine Learning
      8. Security
      9. Software Engineering
      10. Tools & IDEs
      11. Web Design
      12. Web Development
      13. Web Security
      14. Programming Languages
        • PHP
        • JavaScript
      Featured

      Next.js PWA offline capability with Service Worker, no extra package

      August 10, 2025
      Recent

      Next.js PWA offline capability with Service Worker, no extra package

      August 10, 2025

      spatie/laravel-flare

      August 9, 2025

      Establishing Consistent Data Foundations with Laravel’s Database Population System

      August 8, 2025
    • Operating Systems
      1. Windows
      2. Linux
      3. macOS
      Featured

      Windows 11 Copilot gets free access to GPT-5 Thinking, reduced rate limits than ChatGPT Free

      August 10, 2025
      Recent

      Windows 11 Copilot gets free access to GPT-5 Thinking, reduced rate limits than ChatGPT Free

      August 10, 2025

      Best Architecture AI Rendering Platform: 6 Tools Tested

      August 10, 2025

      Microsoft won’t kill off Chromium Edge and PWAs on Windows 10 until October 2028

      August 10, 2025
    • Learning Resources
      • Books
      • Cheatsheets
      • Tutorials & Guides
    Home»Development»Machine Learning»From 100,000 to Under 500 Labels: How Google AI Cuts LLM Training Data by Orders of Magnitude

    From 100,000 to Under 500 Labels: How Google AI Cuts LLM Training Data by Orders of Magnitude

    August 10, 2025

    Google Research has unveiled a groundbreaking method for fine-tuning large language models (LLMs) that slashes the amount of required training data by up to 10,000x, while maintaining or even improving model quality. This approach centers on active learning and focusing expert labeling efforts on the most informative examples—the “boundary cases” where model uncertainty peaks.

    The Traditional Bottleneck

    Fine-tuning LLMs for tasks demanding deep contextual and cultural understanding—like ad content safety or moderation—has typically required massive, high-quality labeled datasets. Most data is benign, meaning that for policy violation detection, only a small fraction of examples matter, driving up the cost and complexity of data curation. Standard methods also struggle to keep up when policies or problematic patterns shift, necessitating expensive retraining.

    Google’s Active Learning Breakthrough

    How It Works:

    • LLM-as-Scout: The LLM is used to scan a vast corpus (hundreds of billions of examples) and identify cases it’s least certain about.
    • Targeted Expert Labeling: Instead of labeling thousands of random examples, human experts only annotate those borderline, confusing items.
    • Iterative Curation: This process repeats, with each batch of new “problematic” examples informed by the latest model’s confusion points.
    • Rapid Convergence: Models are fine-tuned in multiple rounds, and the iteration continues until the model’s output aligns closely with expert judgment—measured by Cohen’s Kappa, which compares agreement between annotators beyond chance.
    Image source: https://research.google/blog/achieving-10000x-training-data-reduction-with-high-fidelity-labels/

    Impact:

    • Data Needs Plummet: In experiments with Gemini Nano-1 and Nano-2 models, alignment with human experts reached parity or better using 250–450 well-chosen examples rather than ~100,000 random crowdsourced labels—a reduction of three to four orders of magnitude.
    • Model Quality Rises: For more complex tasks and larger models, performance improvements reached 55–65% over baseline, demonstrating more reliable alignment with policy experts.
    • Label Efficiency: For reliable gains using tiny datasets, high label quality was consistently necessary (Cohen’s Kappa > 0.8).

    Why It Matters

    This approach flips the traditional paradigm. Rather than drowning models in vast pools of noisy, redundant data, it leverages both LLMs’ ability to identify ambiguous cases and the domain expertise of human annotators where their input is most valuable. The benefits are profound:

    • Cost Reduction: Vastly fewer examples to label, dramatically lowering labor and capital expenditure.
    • Faster Updates: The ability to retrain models on a handful of examples makes adaptation to new abuse patterns, policy changes, or domain shifts rapid and feasible.
    • Societal Impact: Enhanced capacity for contextual and cultural understanding increases the safety and reliability of automated systems handling sensitive content.

    In Summary

    Google’s new methodology enables LLM fine-tuning on complex, evolving tasks with just hundreds (not hundreds of thousands) of targeted, high-fidelity labels—ushering in far leaner, more agile, and cost-effective model development.


    Check out the technical article from Google blog. Feel free to check out our GitHub Page for Tutorials, Codes and Notebooks. Also, feel free to follow us on Twitter and don’t forget to join our 100k+ ML SubReddit and Subscribe to our Newsletter.

    🇬 Star us on GitHub
    🇷 Join our ML Subreddit
    🇸 Sponsor us

    The post From 100,000 to Under 500 Labels: How Google AI Cuts LLM Training Data by Orders of Magnitude appeared first on MarkTechPost.

    Source: Read More 

    Facebook Twitter Reddit Email Copy Link
    Previous ArticleUsing RouteLLM to Optimize LLM Usage
    Next Article AI Agent Trends of 2025: A Transformative Landscape

    Related Posts

    Machine Learning

    How to Evaluate Jailbreak Methods: A Case Study with the StrongREJECT Benchmark

    August 10, 2025
    Machine Learning

    AI Agent Trends of 2025: A Transformative Landscape

    August 10, 2025
    Leave A Reply Cancel Reply

    For security, use of Google's reCAPTCHA service is required which is subject to the Google Privacy Policy and Terms of Use.

    Continue Reading

    Key Metrics That Can Make or Break Your Startup

    Development

    CVE-2025-35005 – Microhard BulletLTE-NA2 and IPn4Gii-NA2 AT+MFMAC Command Injection Vulnerability

    Common Vulnerabilities and Exposures (CVEs)

    ALPHAONE: A Universal Test-Time Framework for Modulating Reasoning in AI Models

    Machine Learning

    Enhance client experience with a custom, all-in-one platform

    Web Development

    Highlights

    ChatGPT can now talk nerdy to you – plus more personalities and other upgrades beyond GPT-5

    August 7, 2025

    GPT-5 is a big deal, but don’t underestimate the power of these smaller feature upgrades…

    VMware fixes four ESXi zero-day bugs exploited at Pwn2Own Berlin

    July 17, 2025

    Black Myth: Wukong is coming to Xbox exactly one year after launching on PlayStation

    June 6, 2025

    GPT-5 is Coming: Revolutionizing Software Testing

    July 22, 2025
    © DevStackTips 2025. All rights reserved.
    • Contact
    • Privacy Policy

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