Conference Presentation, Keynote
Keynote by Karim Beguir, Co-Founder and CEO InstaDeep | RAISE Summit 2024 | Paris
- Industry is projected to enter an "AI super cycle" characterized by exponential data growth, exemplified by genomic sequencing costs dropping from "hundreds of millions" in the early 2000s to "$100," and raw compute for ML tasks expanding by "10x every six months" and "100x every year."
- Hardware efficiency is expected to improve by "1,000x" in GPU efficiency per watt over the "next 15 years" according to TSMC, while visual AI task efficiency requires "two times less compute every 16 months."
- Future AI systems will evolve from single-agent Q&A to collections of collaborating smart agents interacting with human experts, moving beyond text to handling "multiple modalities of data" and performing physical actions.
- AI acceleration will significantly impact workflows, with coding assistance delivering "two times more efficiency" and AI tools like "dpcb.ai" already driving hardware design improvements for entities such as Google and Nvidia.
- Companies are advised to leverage three core assets—data, domain expertise, and simulation capabilities—to create a competitive "virtuous loop" that mimics Tesla's iterative model of human drivers teaching models, thereby "future-proofing" organizations.
- The overarching outlook predicts a "crazy wall of exponential progress" driven by AI accelerating itself, where entities that utilize AI agents as a competitive magnifier are expected to "win very big" in the "near future."