Science Fiction To Science Fact: From The Stargate to the Edge of the World

When I heard that Microsoft and OpenAI were building Stargate, I re-read Arthur C Clark’s A Space Odyssey book. You will probably have seen parts of possibly one of the best movies ever created, but the book is great, too. While it all goes a bit strange at the end, the movie is as fresh to watch these days as it was when it was released. In fact, with the rise of AI, it is probably more relevant:

My favouriate part of the movie and book is, of course, the part with HAL (a HAL 9000 series AI computer), and where HAL thinks that the crew members are risking the mission. Dr David Bowman (“Dave”) — the last survivor on the spaceship — wins in the end and manages to switch HAL off. Overall, HAL plays a greater role in the film than in the book. After the switch-off, Dave finds out that the mission was actually to travel to Jupiter and investigate the source of the signals being sent from the moon. As part of this journey, the Stargate appears, and where it all goes a bit strange.

Basically, the story involves a strange black monolith (TMA-1 — Tycho Magnetic Anomaly One) on the moon that has been sending signals back to Jupiter, and where the crew have a mission to investigate it. In the final part of the book, Dave finds TMA-2 and which becomes a Stargate. This is basically a wormhole into another galaxy. And, so, while the film contains strange sequences of colour, the book vividly describes the sights of travelling through a wormhole in space:

The Stargate to the edge

And, now, Microsoft and OpenAI are investing $100 billion into an AI infrastructure and building the largest data centre ever. This will be full of GPUs with massive computing power and will require a massive amount of electricity. For example, at the current time, it requires the whole charge of a mobile phone to generate a DALLE-3 image. The core of the Internet will be built within the centralised infrastructure.

But what about the edge? Couldn’t we create and AI infrastructure that was located near the demand from users? For this, Cloudflare has just enabled a massive global GPU network and deployed in more than 150 cities. This uses the LLM builder Hugging Face’s multimodel deployment hub and is deployed as a serverless cloud solution. It includes per-trained models that apply to specific domains. These models will aim to push data to where it is required, and also to keep data secure.

The development aims to support customer integration with the Hugging Face open-source models, and this supports improved inference tasks:

Conclusions

The Internet perhaps moves into its end game, and building AI into its infrastructure. Will this happen at the core and be centralised — as with Stargate — or distributed and placed at the edge of the Cloud? We are certainly moving towards a singlarity, and where the machine’s intelligence will soon exceed human intelligence. The investment in GPUs is massive at the current time.