GAEA Talks

Forrester technology and innovation forums (Austin, London, New York). Promo code "GAEATECH" for 10% off. Visit https://forrester.com/events/#tech to book your place.
This week on GAEA Talks, Graeme Scott sits down with Dr Ravinder Singh, Head of Digital and Systems within the UK Cabinet Office Government Commercial Function, and one of the most authoritative voices in Britain on how governments and enterprises should actually build, adopt and govern AI.Ravinder leads the digital, systems and emerging technology work for one of the most consequential functions in Whitehall, and oversees the Government Commercial College - at over one hundred and six thousand users, the largest learning platform in the country after the Open University. Before his current role he was a Consulting Technical Architect at the Government Digital Service. His path into the civil service came after a private sector career at J.P. Morgan, HSBC, Credit Suisse, Accenture, 3i Infotech and Shell Oil. He holds a PhD from King's College London, arrived in the UK on the Highly Skilled Migrant Programme in 2004, and had already spent years as a civil servant in India, where he built the first Indian-languages word processor across seventeen official languages.Filmed in our new London studio, this is one of the most useful, calm and internationally-informed conversations on AI adoption that GAEA Talks has recorded. Ravinder cuts through the noise with a clarity that only comes from having built systems inside global banks, inside the private sector, inside Whitehall and across two countries. The line at the centre of the episode reframes one of the most misused phrases in the industry. Not human in the loop, but human in the lead. Machine learning has to be taught, guided and trained. Trust is earned iteration by iteration. Intelligence comes later. Everything downstream of that principle changes when you accept it.About Dr Ravinder Singh:Dr Ravinder Singh is Head of Digital and Systems within the UK Cabinet Office Government Commercial Function, where he leads the emerging technology, AI, machine learning, blockchain, IoT and quantum computing programmes and oversees the Government Commercial College with over 106,000 users. He was previously a Consulting Technical Architect at the Government Digital Service. Before joining the civil service he spent his career in global financial services and industry, including senior roles at J.P. Morgan, HSBC, Credit Suisse, Accenture, 3i Infotech and Shell Oil, delivering large-scale digital transformation and complex technology programmes. He holds a PhD from King's College London, arrived in the UK in 2004 on the Highly Skilled Migrant Programme, and previously worked as a civil servant in India, where he built the first Indian-languages word processor across seventeen official languages, holds two software copyrights and received a national award for the Punjabi spellchecking algorithm.The views expressed in this conversation are Dr Singh's own and do not represent the official position of the UK Cabinet Office or His Majesty's Government.

Graeme Scott sits down with Alex Whedon, Co-Founder and CTO of Subquadratic and former Head of Generative AI at Tribe AI, for one of the most first-principles conversations GAEA Talks has recorded.
The entire AI industry, Alex argues, is downstream from a single algorithm - the transformer - and its two fundamental flaws are quietly capping what AI can do. He breaks down quadratic compute scaling in plain terms (10x the input, 100x the compute), the memory wall where context can cost more than the model itself, and how Subquadratic's linear-scaling architecture claims to cut compute by up to 1,000x at extreme context lengths without sacrificing quality. Along the way he makes a bracing case that electricity, water, minerals and capital - not clever engineering - are the real limits on AI's growth, that the industry is being far too stingy with tokens, and that efficiency isn't a nice-to-have but an inevitability.
In this episode:
Why the whole AI space is downstream from one algorithm
Quadratic scaling explained simply - 10x the input, 100x the compute
The memory wall, where context can need more memory than the model
What "subquadratic" means, and why linear scaling is the unlock
A claimed ~1,000x compute reduction at 12 million tokens
Why transformers are the worst fit for data-heavy enterprise work
The real constraints on AI: electricity, water, minerals and capital
Why we're "too stingy with the tokens"
The DeepSeek lesson the incumbents ignored
Why first to market is rarely best

About Alex Whedon: Co-Founder and CTO of Subquadratic, which emerged from stealth in May 2026 with $29M in seed funding and SubQ - described as the first frontier model built on a fully subquadratic Sparse Attention architecture, with a research context window of up to 12 million tokens. Previously Head of Generative AI at Tribe AI, leading 40+ enterprise implementations for companies including Anthropic, New Relic and Mars, and earlier an engineer at Meta and Instagram.
GAEA Talks is the enterprise AI podcast for leaders navigating the age of artificial intelligence. New conversations every week.

This week on GAEA Talks, Graeme Scott sits down with Davood Shamsi, Director of AI at J.P. Morgan Chase, Stanford-trained mathematician, former Apple language model lead, and co-author of the forthcoming O'Reilly book From Models to Money.
Filmed in our new London studio, Davood lays out the framework at the heart of his book. He explains why the vast majority of Gen AI pilots fail to produce real ROI, why so many enterprises are quietly hosting "zombie pilots" that nobody wants to kill, and why measuring an efficiency pilot the same way as a strategic bet is one of the biggest mistakes leaders are making right now. He then takes us inside the elegant simplicity of the transformer architecture, explains why data centres are quietly making electricity cheaper for consumers, and walks through the tribal-knowledge shift that will change the value of long-tenured employees inside every large organisation.
Topics covered:
The three questions every AI pilot must answer
The zombie pilot problem inside large enterprises
Efficiency vs compounding vs strategic bet pilots - and why measurement has to change
The Amazon Just Walk Out and IBM Watson lessons
Privacy-first AI - what Apple's approach still teaches every industry
The IPA thesis - why transformers are elegantly simple
Why data centres are quietly making electricity cheaper
The tribal knowledge shift and the future value of long tenure
The price of anarchy and how AI could close the gap in every organisation

This week on GAEA Talks, Graeme Scott sits down with James Palles-Dimmock, CEO of Quantum Motion, Cambridge-trained physicist, and one of the sharpest voices in the world on how quantum computing will actually reach useful scale.
Filmed in the new London studio, this is one of the most technically substantive and hype-free conversations on quantum computing GAEA Talks has recorded. James's argument is simple and radical. Quantum computing will not scale by adding qubits one at a time in a university lab. It will scale by riding the only industrial process that can hit millions or billions of units, which is CMOS. That is the bet Quantum Motion is making, backed by over one hundred and sixty million dollars of funding, a deployed machine at the National Quantum Computing Centre and a leading position in DARPA's Quantum Benchmarking Initiative.
Topics covered:
What a quantum computer actually is (and what it is not)
Why silicon spin qubits are the only credible route to million-qubit scale
The AI and quantum crossover - why quantum is a data generator for AI, not a competitor
Why AI's real limitation is data and world models, not architecture
The Landauer limit and reversible computing
Steve Jobs's "bicycle of the mind" and specific vs general AI
The scientific method in one line - "try to make mistakes as quickly as possible"
Sovereign AI as resilience, not autarky
The Quantum Motion roadmap to a utility-scale quantum computer by 2032 to 2033
Why the real breakthroughs will come from a fourteen year old in her bedroom, not a warehouse

This week on GAEA Talks, Graeme Scott sits down with Professor Michael Jacobides - Sir Donald Gordon Professor of Entrepreneurship and Innovation at London Business School, academic advisor at the BCG Henderson Institute, and one of the most respected minds anywhere on how AI is reshaping the structure of the modern enterprise.
Filmed live in our new London studio, this is a masterclass in cutting through the current AI hype. Michael explains why the current wave of Gen AI is built on capital markets expectations rather than business outcomes, why the only people making real money from AI today are the picks and shovels, and why "AI first" is one of the silliest strategic frames in the market. He walks through his white rabbit and EBITDA elephant framework, the difference between productivity gains and value proposition disruption, the Chinese pragmatism versus US "race to become God" divide, and why the historical link between US academia and industry that produced everything from the Ethernet to the transformer paper is now being deliberately torn apart.
Topics covered:
Why AI needs to be analysed as an ecosystem, not a technology
The capital markets problem in Gen AI valuations
The picks and shovels reality - who is actually making money today
Gen AI as a mass persuasion technology, not a truth technology
The white rabbit and EBITDA elephant framework
Why value proposition disruption matters more than productivity
The Chinese pragmatism versus US "race to become God" divide
Why the DARPA-to-transformer academic pipeline is under threat
Teddy Roosevelt's rule - feet on the ground and eyes on the stars