Speed-of-Light Computing Could Get Us to AGI Very Fast

Quick Overview

Photonic computing, which uses light particles (photons) instead of electricity for processing, promises substantial speed, efficiency, and multichannel advantages over current electronic chips, but it currently faces significant drawbacks in storage and universality, meaning that while Chinese scientists achieved a 12.5 GHz optical AI chip, the technology still requires electronic components for memory storage and needs more research to overcome limitations like error rates before it can fully replace silicon designs or achieve AGI.

Key Points: Photonic computing utilizes photons instead of electrons, offering speed advantages where calculations occur at the speed of light (12.5 GHz demonstrated by a Tsinghua chip vs. 100-500 ns for electronic chips). Key advantages of photonic computing are speed, efficiency (requiring only 10% of the power of electronic equivalents), and multichannel capability via different frequency bands. Major drawbacks include the inability to store information using photons, requiring supplementary electronic components for storage, and a lack of universality for general logical operations. Chinese scientists at Tsinghua University created the world's fastest optical AI chip running at 12.5 GHz, serving as a strong benchmark, although the presenter rates the hype surrounding this as a 5/10. US startups like Lightelligence and Qant are also developing photonic processors, claiming up to 50x performance improvements and significant energy reduction (up to 90%). The technology is already being used in aerospace, biomedicine, and finance, but achieving the precision and reliability of established electronic methods remains a challenge. The presenter concludes that while photonic computing is promising, more research is needed to address error rates and universality issues before it can completely supplant electronics.

Context: This video provides a scientific news update, hosted by Sabine Hossenfelder, discussing the emerging field of photonic computing—the use of light (photons) rather than electrical current to perform computations. The discussion centers on recent progress, particularly developments in China and the US, weighing the touted benefits like extreme speed and low energy consumption against inherent technological limitations like data storage and operational universality.