Crème de la LLM: How language models are enhancing machine learning applications

This article first appeared in Sifted’s Daily newsletter, sign up here.

London Tech Week ended in style on Friday, at an unofficial side event hosted by AI investor Nathan Benaich’s VC firm Air Street Capital.

The Research and Applied AI Summit (RAAIS), now in its eighth edition, hosted founders and execs from some of the UK’s leading AI scaleups like self-driving car company Wayve, B2B video creation tool Synthesia and enterprise AI platform InstaDeep, which was acquired by BioNTech in 2023 for €500m.

Benaich told Sifted that, over 10 years of running the event, the scope of problems being solved with AI has widened dramatically.

“You have biotech, defence, enterprise automation, image, video — all the modalities,” he said. “The diversity of types of problems people are working on is way bigger.”

And, in a week where European tech was set abuzz by Paris-based GenAI startup Mistral raising €600m at a €5.8bn valuation, one big theme was the rollout of large language model (LLM) technology across other, more established forms of machine learning applications.

“What I think is pretty cool is all these companies that have worked on more ‘traditional’ machine learning systems are seeing very interesting ways to add LLMs to their stack,” said Benaich.

“Whether that’s Wayve with interpreting driving scenes, or Synthesia with better adaptability of its avatars… It’s juicing existing systems and making them more usable.”

InstaDeep cofounder and CEO Karim Begur described how his company had been applying open-source LLMs, such as Meta’s Llama, to masses of biological data to create a ChatNT (nucleotide transformer) that could open up new avenues of research over the next 3-5 years.

“I believe we will see revolutionary results in the future,” he said. “It’s a very exciting time for AI in biology.”

Wayve —which began life as a company largely focused on vision models — last year released a model called LINGO-1, which combines vision and language AI technology to deliver enhanced learning abilities.

“Over the last five years, it’s been all about language,” said cofounder and CEO Alex Kendall.

Another message that rang through loud and clear was the need for patience when building impactful AI companies.

“Now everybody’s starting companies kind of hoping or expecting instant success,” said Benaich. “Many of the stories here, whether it’s Wayve, or Synthesia, or the AlphaGo story — they took ages. The sort of meta-message is: be patient and persistent.”