Roundtable, Conference Presentation
'GenAI at Scale Measuring Enterprise Outcomes in the AI-First World' | RAISE Summit 2024 | Paris
RAISE SummitFrançois Candelon, Yuriy Nevmyvaka, Yannis Daubin, Harry Grieve, Jeremy Harroch, Antoine Courret
- Morgan Stanley intends to accelerate the deployment of machine learning activities in banking, currently operating dozens of production use cases with a 20-person research team planning to establish a PhD recruitment pipeline within one to three years.
- François Candelot anticipates leaving BCG in May to join the private equity firm 7-2, aiming to invest in profitable AI startups and help portfolio companies leverage technology while predicting that Gen AI will significantly impact consulting.
- Market analysis suggests the current Gen AI landscape is at an early stage with limited large-scale implementation despite high interest, driven by a necessity for clearer data platforms to overcome adoption blockers.
- Competitive advantage is expected to derive from the rate of learning and technological adoption rather than pure technological capability, with experts noting that 10% of science could be exceeded by AI within a current effort distribution of 10% technology, 20% infrastructure, and 70% humans.
- Future predictions include the potential emergence of consumer brain-machine interfaces within a decade, raising concerns about government capture of artificial neurons and the critical need to preserve open-source nature to prevent dystopian outcomes.
- Strategic success is viewed as dependent on having a single source of data, strong leadership, and significant internal adoption, as lack of these factors renders even advanced use cases valueless.
- Regulatory frameworks are projected to become increasingly important in the coming years, potentially requiring a trade-off between experimentation and safety, with hopes for open regulation that facilitates interaction while allowing diverse entities to excel.
- The immediate focus for research and development is expected to shift toward repurposing large language models and diffusion models to address the underutilized modality of tabular time-series data.