AI Research Foundry London Launches with DeepMind, Microsoft and Imperial — EdTech Innovation Hub

AI Research Foundry London launches with Imperial, DeepMind and Microsoft researchers

The new AI research community will bring academia, technology companies and startups together for detailed paper discussions, with its first event on 22 October already sold out

AI Research Foundry London is launching a new paper discussion community, bringing together researchers from Imperial College London, Google DeepMind, Microsoft AI and Pleias for its inaugural session on Thursday 22 October 2026.

Founded by Shubham Jain, Lead Machine Learning Scientist at Sense Street, the initiative is designed to create a space for researchers working across academia, major technology companies and startups to examine shared research in depth.

The first event, titled Data Is All You Need, takes place at Pillar VC in King's Cross, London, from 6pm to 8:30pm. Attendance is free, although registration has already sold out.

Jain explained the thinking behind the initiative in a LinkedIn post: "London has great communities for builders. I couldn't find one where researchers go deep on a shared paper."

Imperial, DeepMind, Microsoft and Pleias join the first session

The opening discussion focuses on data creation, curation and scaling, examining the role training data plays in improving AI systems.

Four researchers will contribute to the session, bringing perspectives from academic research and commercial AI development.

Joshua Placidi, a PhD student in Imperial College London's Brain and Behaviour Lab, will discuss sequence composition in electronic health record (EHR) foundation model pretraining. His research includes scaling self-supervised learning across medical data using generative foundation models.

Google DeepMind Senior Research Scientist Yash Bhalgat, who works on Gemini Omni and previously served as lead captain for Omni 1.5 pretraining, is also joining the session. His individual talk topic has yet to be confirmed.

Microsoft AI's Maximilian Croci will discuss MAI-Thinking-1 and its development, drawing on his work with pretraining data, including mathematics, science and code sourced from the web.

Pierre-Carl Langlais, Co-founder and CTO of French-German AI lab Pleias, completes the speaker lineup. His session will examine a fully synthetic, single-stage approach to training large language models (LLMs).

Why training data is the opening focus

The first session takes its starting point from research published in September by Dwarkesh Patel and Jerry Han, titled Pretraining progress is mostly coming from data.

Their experiments compared improvements in publicly available model training methods and datasets between 2019 and 2025.

At a training compute budget of 10¹⁹ floating-point operations (FLOPs), the researchers found that data improvements delivered a 12-fold gain in compute efficiency, compared with a 3.7-fold gain from model improvements. In that experimental setting, the gains attributable to data were 3.24 times greater.

The authors caution that their experiments were conducted at a relatively small scale and may not capture the full contribution of model improvements at the frontier. They also did not investigate whether synthetic data could effectively expand training datasets without compromising performance.

The topic provides the starting point for the Foundry's discussion format, which begins with a shared paper before opening the conversation to questions and wider examination. Participants are encouraged to read the research beforehand, with summaries provided ahead of sessions.

Pillar VC is hosting the inaugural event, with sponsorship from the UK Dynamism Fund through a Sparks Grant, according to Jain's announcement.

AI Research Foundry London plans to meet approximately once a month or every six weeks, with each session centered on one research theme and a selection of papers. Participation will remain free, with limited places and curated attendance.

Registration for the 22 October session is now closed after selling out.

ETIH Innovation Awards 2027 open for entries

The ETIH Innovation Awards 2027 are now open for entries, with 26 categories recognising companies, institutions, projects and individuals making a measurable impact across education technology. The global awards span AI in education, digital learning, higher education, skills and workforce, inclusion, education operations, evidence and impact, with entries managed through ETIH’s new online awards platform.

Entries are open worldwide, with the early-bird deadline on Friday 18 December 2026 and the final entry deadline on Friday 26 February 2027 at 23:59 UK time.

Explore the 2027 categories and judging criteria, read the How to Enter guide or start an entry through the ETIH Innovation Awards 2027 platform.