Ian Ballantyne

On 20th November we hit a milestone For those of you that know I've been working on improving Localised Ai Memory. It's my belief that we should own our data and not rely on software companies as they introduce an element of counter party risk to security and privacy. I've been working with Jack, a long time friend and business partner who has been working relentlessly on AutoMem (which is opensource and free for anyone to use) It is now the world's best-performing long-term memory system for AI assistants. On November 20, 2025, AutoMem achieved 90.38% accuracy on the LoCoMo benchmark (ACL 2024), beating the previous state-of-the-art by 2.14 points. This is the academic standard for long-term conversational memory. This makes AutoMem the highest-performing memory system in the world, beating: - CORE (heysol.ai): 88.24% (previous SOTA) - OpenAI's baseline: 39% Key breakthroughs: - Multi-hop bridge discovery - Finds connecting memories across conversation threads - Temporal alignment scoring - Understands time-aware queries ("what happened last year?") - 9-component hybrid search - Semantic + lexical + graph + temporal signals - 100% complex reasoning - Perfect score on multi-step reasoning tasks Full benchmark results → Why AutoMem Exists: Your AI forgets everything between sessions. RAG dumps similar documents. Vector databases match keywords but miss meaning. None of them learn. AutoMem gives AI assistants the ability to remember, connect, and evolve their understanding over time—just like human long-term memory. What Makes AutoMem State-of-the-Art AutoMem is a graph-vector memory service built on peer-reviewed neuroscience and validated by academic benchmarks: - Stores memories with metadata, importance scores, temporal context, and semantic embeddings - Recalls via 9-component hybrid search (vector + keyword + graph + temporal + lexical) - Connects memories with 11 typed relationships (RELATES_TO, LEADS_TO, CONTRADICTS, etc.) - Learns through automatic entity extraction, pattern detection, and consolidation cycles - Reasons with multi-hop bridge discovery—finds connecting memories across conversation threads - Performs with sub-100ms recall across thousands of memories AutoMem implements breakthroughs from: HippoRAG 2 (Ohio State, 2025): Graph-vector hybrid matches human associative memory A-MEM (2025): Dynamic memory organization with Zettelkasten-inspired clustering MELODI (DeepMind, 2024): 8x compression without quality loss via gist representations ReadAgent (DeepMind, 2024): 20x context extension through episodic memory Enjoy https://lnkd.in/e4rp88ZG

4

Hugely excited for our announcement tomorrow at GDC Festival of Gaming! The debate about AI in games has become ever more polarised - with people either violently rejecting it, or blindly succumbing to our new big tech overlords. At Meaning Machine we've always tried to hold a position of nuance. Everything bad you say about AI? We're almost certainly agree with it. But equally we also believe that generative technology has the potential to be truly transformative for interactive entertainment - if only the human author can seize control of its power. Our talk at GDC, in partnership with University of Bristol's Richard Cole, will provide nuance by putting the player at the heart of the debate. It delivers some of the very first insight into what players ACTUALLY feel about AI in a game - when they experience it firsthand. Going beyond marketing surveys, culture wars, and hype bubbles. Can't wait to share the findings with the world!

51

1 Comment

“If you’re not first, you’re last.” – Ricky Bobby  (…and probably your next AI-powered competitor) AI isn’t in R&D anymore — it’s in production. I joined Conor Bronsdon on the Chain of Thought podcast to talk about what it means to shift from playing with AI in a sandbox … to building the future others will copy. 🎧 Give it a listen — then ask yourself: Am I moving fast enough? #AIInProduction #AgenticAI #AILeadership #ChainOfThought

62

2 Comments

We’re right in the middle of interviews for our next cohort—and it’s been incredibly busy in the best possible way. Applications have been hugely positive this year, with real growth in interest. What’s especially encouraging is that, while much of our outreach has focused on the North and North West, we’re also seeing strong engagement from the South of England. Over the past few weeks, I’ve seen exceptional work, genuine drive, and exciting talent. Even more valuable have been the conversations about games—and making games—that continue to fuel my own commitment to delivering what we set out to achieve at TGA. From the beginning, I made a simple decision: everyone who applies deserves the opportunity for an interview. Where possible, this is in person; when travel is a barrier, we meet remotely. For many applicants, this is their first formal step toward a career in the games industry, and they deserve time to ask questions, share ambitions, and explore whether this path is right for them. When people are being asked to invest significant money—and take on years of student debt—decisions should be informed, personal, and rooted in real conversation, not just paperwork. Portfolios are central to this process. They allow applicants to demonstrate skill, passion, and potential, and the discussions around them often reveal far more than what’s initially submitted. Just as importantly, the interview ensures that the decision is mutual. Even when TGA isn’t the right pathway, no one should leave without feedback, guidance, and a clearer sense of direction. There are many routes into the games industry, and different journeys can still lead to the same goal. What matters most to us is simple: every applicant deserves time, respect, and a genuine conversation before making such an important decision about their future. That belief sits at the heart of how we interview, how we review portfolios, and how we support people—whether they join TGA or take a different path into the industry. Because meaningful education—and lasting careers in games—begin with being seen, heard, and understood. #GamesEducation #HigherEducation #StudentSuccess #FutureTalent #EarlyCareers

42

2 Comments

Organoid computers, or biocomputers, are an emerging technology using lab-grown, 3D human brain cell clusters (organoids) as biological processors to perform computing tasks. https://lnkd.in/ew_58gmV NB: One of the UK's biggest car park operators, National Car Parks (NCP), has gone into administration. It's thought that around 682 jobs could be at risk as well as the potential closure of 340 sites across the UK. https://lnkd.in/e2xP3Tn9 Attention: PwC .... 😈

The new programming language is human. Thoughts? Today, with Jensen Huang, Sir Keir Starmer PM and Poppy Gustafsson together at #LondonTechWeek #NVIDIA Department for Business and Trade Department for Science, Innovation and Technology Department for Culture, Media and Sport Charismatic.ai

25

6 Comments

The impact of new tools is often transformational, and the emergence of generative AI is breaking pre-conceptions of how future media will be formed. Right now AI image and video generation is rarely achieved interactively, but takes many seconds, minutes or even hours and often through extended trial and error sequences of text prompting. Last week #RavensbourneUniversity (a focused creative media university in London) welcomed me to speak on recent results in my roles for Edinburgh Napier University, 3FINERY LTD and Cobra Simulation Limited on Delivering Interactive Generative AI for Everyone. Talk slides recording now posted on YouTube https://lnkd.in/gsBHgGpr

29

1 Comment

The policy team at Ukie (or more properly IES) has released its #manifesto in the run-up to the Scottish Parliament elections. It's a great piece of work, very strong and focused on growing understanding and recognition of the games sector, introducing funding and a new focus on education and skills. Sound familiar? It's entirely aligned with the five key recommendations of our own #LevelUp: Scotland's Games Action Plan, which we published in January. However, LevelUp is already with the Business & Employment minister, with a pledge to attend the next cross-party group meeting at the Scottish Parliment for a line-by-line review. That date is awaiting final confirmation. Find the manifesto and read more here: (And if you've not yet read, or left a statement of support for #LevelUp, there's still time. Full details on the SGN website).

11

I was honored to introduce a standout session at the UCL AI Festival this morning, featuring the incredibly impressive Lourdes Agapito (Co-founder of Synthesia and Professor at UCL) and the brilliant Rafah El-Khatib who works with the brightest and best frontier AI companies in the UK at (NVIDIA). Their discussion on the "Research to Reality" leap offered a look at how foundational 3D vision research transitions into a global enterprise platform. It was a particular point of pride to highlight the role of the NVIDIA Inception programme during this introduction. Seeing a company scale from its origins in academia to a platform serving 90% of the Fortune 100 underscores the impact of providing the right computational and ecosystem support at the right time. Key Themes from the Discussion: * The Inception Catalyst: They reflected on the importance of the Inception programme in bridging the gap between a university prototype and a world-leading $4B unicorn. * A Shared Vision: A central takeaway was the necessity for founders to maintain a common vision, balancing the curiosity-driven nature of academic research with the practical and scalable, product-driven requirements of the market. * The UK Ecosystem: Both speakers identified London and the broader UK AI landscape as a unique environment for building deep-tech. Drawing inspiration from the DeepMind journey, they noted that the local talent pool remains a significant strategic advantage. Synthesia, Wayve, Elevenlabs are all in this next cohort of superstars made in the UK. * Ethics and Responsibility: The conversation emphasised that ethical pillars are not elective. Safeguarding synthetic media remains a core priority for Synthesia As the industry moves toward interactive audio-visual agents, they explored the technical challenges of real-time 3D vision, with innovation moving from long term research to product. My thanks to Rafah and Lourdes for sharing such inspirational insights during the UCL AI Festival. Any aspiring founders in the room can see their work continues to set a high standard for how we bridge the gap between world-class academia and industrial excellence. Howard Wright Tobias Halloran Matteo Sarcinella Chiara Refäuter Graça Carvalho Patrick Gilday Hemesh Prag #UCLAIFestival #AI #DeepTech #NVIDIAInception #Synthesia #ComputerVision #UKInnovation #UCL #DigitalHumans #Nvidiaforstartups

180

5 Comments

If you’re building advanced AI and not thinking about how it could be weaponised, you’re being naive. I’ve spent the past few years developing AI systems in: • Context-driven language models • Voice modelling and lip-sync • Image manipulation and synthesis • Real-time media automation And here’s the uncomfortable truth: The line between innovation and exploitation is thinner than most executives realise. When you build models that can replicate tone, reconstruct voice, simulate identity and automate contextual decision-making — you quickly understand how easily the same capability could be used to manipulate, deceive or infiltrate. AI doesn’t just improve workflows. It industrialises capability. The myth: “AI risk is theoretical.” It isn’t. If you know how these systems work, you know: • Identity is now programmable • Trust signals can be fabricated • Expertise can be compressed into prompts • Automation removes skill barriers We’ve effectively lowered the entry point for sophisticated digital manipulation. That genie isn’t going back in the bottle. Here’s the part nobody likes to say out loud: Most AI products are being rushed to market without serious architectural security thinking. Innovation is rewarded. Governance is postponed. That’s backwards. When building our own AI systems, we assumed misuse from day one. So we implemented: • Zero training on client assets • Strict model and data isolation • AES-256 encryption by default • Hardened IAM and API rate controls • Full audit traceability • Human approval layers for sensitive synthesis Not because it’s fashionable. Because it’s necessary and yet I am still concerned we have not done enough. AI is not dangerous because it’s intelligent. It’s dangerous because it’s accessible. And the builders carry responsibility. If you’re deploying AI at scale without embedding security at architecture level, you’re not innovating — you’re creating future incident reports. It's a Hard truth. Curious how many AI companies are genuinely designing for abuse scenarios rather than just growth curves. #AI #CyberSecurity #ResponsibleAI #CTO #DigitalRisk

8

https://lnkd.in/ejcYbquZ "Inevitably, game engine company Unity has been looking to both drive and capitalise on the AI boom, with CEO Matthew Bromberg now bullishly highlighting what he sees as the next step in 'democratising' game development." Democratising development == flooding the market with absolute trash. Wealthy individuals never seem interested in making the things that improve our quality of life more accessible, only the technologies that replace skilled workers who earn a living from their expertise. How convenient. The sooner this bubble bursts the better, we don't need AI in every single aspect of our lives, no one is asking for it, no one needs it. The only beneficieries from this revolution are the tiny group of incredibly wealthy individuals sitting atop their mountains of money watching the rest of the world queueing for food banks and struggling to make a living because their jobs are being taken over by AI.

7