Keep pulling the thread on Arvind Krishna.
An internal IBM team of 6,000 software developers using a custom AI coding tool became 45% more productive within four months compared to their peers.
Arvind Krishna predicts that the cost of AI compute will become 1,000 times cheaper over the next five years.
Arvind Krishna asserts that Large Language Models (LLMs) provide a 100x advantage in speed, tuning, and deployability compared to previous bespoke deep learning methods.
Announced commitments for AI data center build-outs total approximately 100 gigawatts, representing a potential $8 trillion in capital expenditures.
Arvind Krishna predicts that AI will likely cause up to 10% job displacement in the total US employment pool over the next two years, concentrated in specific areas.
Arvind Krishna does not believe the current AI market is a bubble, but expects that some of the capital being spent, particularly debt capital, will not generate a return.
The current cost to build and equip a one-gigawatt AI data center is approximately $80 billion.
Consulting firms like Boston Consulting Group and McKinsey estimate the potential annual value created by utility-scale quantum computing to be between $400 billion and $700 billion.
The projected 1000x reduction in AI compute cost over five years will be driven by a 10x improvement from silicon advances, a 10x improvement from new chip designs, and a 10x improvement from software and memory optimizations.
Arvind Krishna predicts the AI era will lead to the creation of one billion new applications, significantly more than the few million developed for the smartphone ecosystem.
Achieving AGI will likely require fusing LLMs with technologies that represent hard knowledge, such as neuro-symbolic AI.
Arvind Krishna believes there is a very low probability, between 0% and 1%, that the current set of known technologies like LLMs will lead to Artificial General Intelligence (AGI).