Yet another startup wants to crack the LLM code but this time using light; optical pioneer Oriole Networks wants to train LLMs 100x faster with a fraction of the power

A mates of years ago, Israeli startup CogniFiber made headlines pinch Deeplight, a fiber-optic cablegram which could, “process analyzable algorithms wrong nan fibre itself earlier nan awesome hits nan terminal.” At nan time, we warned this exertion wouldn’t scope extremity users successful nan adjacent future, and was improbable to look successful laptops aliases smartphones anytime soon.

However, eeNews Embedded is now reporting connected Oriole Networks, a UK-based startup utilizing ray for a different intent - to create businesslike networks of AI chips.

The exertion tin reportedly train LLMs up to 100 times faster than accepted methods while drastically reducing powerfulness usage, and this investigation intends to mitigate nan increasing power depletion of information centers driven by nan accelerated description of AI workloads and nan expanding request for high-performance computing.

20 years of photonics investigation

“Our ambition is to create an ecosystem of photonic networking that tin reshape this manufacture by solving today’s bottlenecks and enabling greater title astatine nan GPU layer. Building connected decades of research, we’re paving nan measurement for faster, much efficient, much sustainable AI," said James Regan, CEO of Oriole Networks.

The company's roots dishonesty successful optical web investigation from University College London (UCL) and Oriole’s unsocial IP is based connected nan activity of founding scientists Professor George Zervas, Alessandro Ottino, and Joshua Benjamin.

The startup has already drawn attraction from a number of investors keen to find solutions for AI's expanding power demands.

With plans to merchandise its early-stage products by 2025, Oriole Networks hopes to reshape AI’s infrastructure by making it faster, much energy-efficient, and yet much sustainable.

eeNews Embedded quotes Ian Hogarth, a personnel of Oriole’s board, and a partner astatine Plural which led nan latest backing round, “Applying 20 years of heavy investigation and learning successful photonics to create a amended AI infrastructure demonstrates really overmuch much invention location is to travel to thief reap nan benefits of this technology," he said.

"The squad down Oriole Networks has proven acquisition successful some institution building and bringing heavy subject to commercialization, creating a basal displacement successful nan creation of next-generation networked systems that will trim latency and slash nan power effect of information centers connected which we now rely.”

More from TechRadar Pro

• Researchers create RGB LED retired of occurrence worldly perovskite
• AI managed to creation a robot that walks - and successful little than 60 seconds
• Liquid metallic RAM is first measurement towards shapeless computing