AI Infra Summit: Val Bercovici Keynote – Tokenomics. Latency. Survival Economics for GenAI

In a recent talk, Val Bercovici reflects on the rapid evolution of AI models, transitioning from generative pre-training to reasoning models, which have surpassed previous limitations. They highlighted the impressive capabilities of these new models, particularly in the context of defeating established benchmarks like the ArtPrize. The discussion emphasized the critical role of memory in AI performance, noting that advancements in memory technology have not kept pace with the demands of AI. The speaker also pointed out the importance of balancing accuracy, cost, and latency in AI inference to ensure profitability. They concluded by encouraging collaboration and exploration of innovative approaches to further advance the field.