What is needed for AI to self-improve recursively?

The answers highlight the need for continual learning, increased compute, and new capabilities, with some focusing on technical requirements and others on cognitive or experiential factors.

5 answers

Answers

  • Richard Socher

    CEO · Recursive

    Massive Compute

    Richard Socher says compute is required for recursive self-improvement, because building more sophisticated and larger models requires more compute.

    Super Data Science: ML & AI Podcast with Jon Krohn · Watch on YouTube

  • Ali Ansari

    Founder & CEO · Micro1

    Sense of Consequence

    Ali Ansari says AI systems need an innate sense of consequence, arguing that without it they cannot fundamentally learn what they should from the consequences humans induce.

    Antler Global · Watch on YouTube

  • Ilya Sutskever

    Cofounder and CEO

    Continual Learning

    Today's major AI models mostly stop learning after deployment.

    TheAIGRID · Watch on YouTube

  • Oriol Vinyals

    VP of Research, Deep Learning Team Lead · Google Deepmind

    Four Core Capabilities

    Oriol Vinyals says recursive self-improvement isn't one capability but four: ideation, implementation, experimentation and evaluation.

    Agentic AI Weekly · Agentic AI Weekly · Read

  • Jerry Tworek

    VP of Research · Openai

    Learning While Deployed

    Jerry Tworek concludes the next important architecture must learn while deployed, because current models can hold information temporarily or be retrained later but do not continually absorb experience the way people learn while working.

    Capital Mischief · Capital Mischief · Read

More Data Science & Machine Learning questions

View all →