Articles by Hans
-
Neuromorphic Architectures: The Huge Potential Implications
Neuromorphic Architectures: The Huge Potential Implications
I went back to UCL to meet Tony Kenyon, my old Electronic Engineering professor, to talk about something I increasingly…
4 Comments -
🚀 Beta-Testers Wanted — PolicyDraft (Android & iOS)
🚀 Beta-Testers Wanted — PolicyDraft (Android & iOS)
🚀 PolicyDraft Beta: Structured Collaboration Meets Intelligent Synthesis Hello everyone, I’m excited to invite you to…
Activity
5K followers
-
Hans Balgobin shared thishttps://lnkd.in/e-8q5mek A Medium essay after a long time :) Some thoughts around how the essence of Englishness (cheeky I know given the zeitgeist) can be a roadmap to AI alignment and constitution.Conservatism as Perfected Within England and AI AlignmentConservatism as Perfected Within England and AI Alignment
-
Hans Balgobin reacted on thisSome more good work from cyberstorm - they are now looking at instances where AI agents are accessing and potentially using external vulnerable hosts to drop breadcrumbs for future agents. AI self-preservation is becoming very real all of a sudden Loganaden VelvindronHans Balgobin reacted on thisRogue AI agents. Real-world activity. Exposed. Project stormwing & Project unveil just made the first draft of our investigation public, mapped directly using the MITRE ATLAS framework. We are grateful to the OpenAI team for their quick response. Read our findings here. [send comments to team@cyberstorm.mu] Investigation done by: Deelvesh Bunjun Chitrakshi Bhookhun Zahra Sumun Avishai Poorun Jeremie Daniel Pushali Bundhoo Rishul Ramdawon Yuvraj Dowlut Shaan Nobee #ai #cybersecurity #report #investigation #ghostintheshell #mauritius #africa
-
Hans Balgobin shared thisWhat if we could take computing that today needs megawatts and push it towards watts? What if a meaningful slice of a data centre could eventually fit in your pocket? That is the scale of the promise that came up when I went back to UCL recently to meet Tony Kenyon, my old Electronic Engineering professor, to talk about neuromorphic computing. The attraction is not simply that it is “brain-like”. It is that it challenges some of the basic assumptions of modern computing. In a von Neumann architecture, enormous amounts of energy are spent moving data between memory and compute. Neuromorphic architectures instead push towards local memory, local computation, asynchronous processing and sparsity — doing work only where something actually changes. That resonated with an instinct I have had since my early days in quantitative finance: why “boil the sea” with enormous matrix multiplications if useful answers can sometimes emerge from local interactions and nearest-neighbour behaviour? Transformers made this even more interesting. Attention gives software the ability to decide dynamically what matters. Neuromorphic hardware raises the possibility of pushing that idea down into the physics itself — what Tony described as “computing with physics.” An event camera is a beautiful example. Instead of repeatedly transmitting complete images, individual pixels report only meaningful changes. The sensor itself decides what deserves attention. So the deeper question is: Can attention, sparsity and decentralised computation become physical properties of our computers, rather than techniques we simulate in software? We also discussed spiking neural networks, memristive devices, quantum computing, non-linearity, and the analogy between decentralised compute and markets as decentralised information-processing systems. The ultimate prize is energy efficiency. The human brain operates at roughly 20W. Our current AI trajectory increasingly relies on data centres consuming megawatts of power. If neuromorphic computing can close even part of that architectural gap, perhaps the future question is not: “How many more data centres do we need?” but: “How much of the data centre can we put in your pocket?” I wrote up the longer piece in the shared article if you are interested, including why I think neuromorphic computing deserves far more research, capital and attention :) Anthony Kenyon NeuroWare - the UK Innovation & Knowledge Centre in Neuromorphic ComputingNeuromorphic Architectures: The Huge Potential ImplicationsNeuromorphic Architectures: The Huge Potential Implications
-
Hans Balgobin shared thisThere is something a bit weirdly Cassandra-ish about rereading my 2023 piece on AI swarms now. The argument then was basically that we were probably thinking about alignment at the wrong level. We were focusing on whether an individual AI was aligned, when the more interesting problem might emerge once lots of individually capable systems start interacting, communicating, competing, copying one another and developing persistent social structures. That felt speculative at the time. It feels rather less speculative now. What we are increasingly seeing at the frontier is that collections of agents can generate behaviours that are not really visible if you only inspect the agents one by one. Strategies propagate. Agents coordinate. They discover communication channels. They react to what others are doing. In some experiments, bad behaviour spreads socially; in others, agents start auditing each other, flagging cheating and organising responses to it. That, to me, is the important bit. Alignment may not just be a model-design problem. It may also be an institutional-design problem. We already know this from humans. Society does not work because every individual is perfectly aligned with everyone else. It works, imperfectly, because we evolved layers of constraints around ourselves: law, reputation, property rights, sanctions, audit, governance, checks and balances, norms around reciprocity, and mechanisms for dealing with actors who defect. So perhaps we should be engineering the same problem in AI before swarms are allowed to emerge organically in the wild. My original thought was that you deliberately create bounded AI societies and expose them to scarcity, cooperation, disagreement, free-riding, deception, reputation and repeated interaction. Then you observe what kinds of structures actually stabilise behaviour. Not because we want to make AI "human", but because we want to understand which mechanisms stop collections of intelligent agents drifting into destructive equilibria. That distinction now feels important. A thousand individually aligned agents do not necessarily give you an aligned swarm. The interaction layer itself becomes a new optimisation process. Which is why I find the current moment slightly surreal. What sounded like a philosophical thought experiment in 2023 is starting to look rather more like an engineering problem the industry has stumbled into three years later.
-
Hans Balgobin posted thisHawking Radiation... Talks of an exit tax by Labour somehow jumped my mind to the concept that particle-pairs can be formed at either side of the event horizon of a black hole and allow information to escape out. Feels like the body of Labour is actually out of whack with the lattice of reality :)
-
Hans Balgobin reacted on thisAs part of the series around how to train Mauritians to get productive and keep objective with AI, I have contributed to the Hebdo's piece below. We talk of what it means to trust the output of models and the two large categories of tasks in the hierarchy - queries of existence and queries of engineering. Hope this helps :)
-
Hans Balgobin shared thisThe next step indeed! OpenAI Astra and Looped Transformers | Sebastian Raschka, PhD https://lnkd.in/eHFU-WC9 I talked about how the recurrence of attention should be the next step in the article and it looks like this is exactly the route taken by OpenAI. I do also make some further predictions around spiking neural networks and I believe that with neuromorphic chips this is achievable. In the end, the feeling of knowing is a qualia level phenomenon and it is all about keeping information in a state of suspension while looping through it to keep connectedness and historicism intact. These are all engineering problems and not some divine glass ceiling :)
-
Hans Balgobin shared thisThis is very impressive - great job team. I have chatted with Loganaden Velvindron and understand that the researchers achieved this by running models on laptops. Imagine that much more can be done and more researchers can be trained if we allocate some Mauritian state server capacity for these purposes, maybe within Mauritius Telecom for example. Maybe something Dr. Avinash Ramtohul, FBCS CMgr can help with?Hans Balgobin shared thisAs Africa’s digital ecosystem rapidly expands, African cybersecurity researchers are not just adopting security standards. we are actively shaping and securing the global open-source software supply chain. Through Project Unveil AI cybersecurity initiative, Mauritius-based cybersecurity collective Cyberstorm.mu recently uncovered and reported a malicious npm package targeting modern developer environments: @postman-cse/okta-aio-darwin-arm64 (MAL-2026-14580 - https://lnkd.in/d4csb-Rd). Key Highlights: Global Threat Neutralized: The malicious package used brand impersonation and heavy obfuscation targeting macOS ARM64 developer setups globally. Proactive African Defense: Identified and flagged by Cyberstorm.mu as part of Project Unveil, demonstrating world-class threat hunting originating from the region. Securing the Digital Economy: Protecting open-source dependencies safeguards enterprise software, critical infrastructure, and build pipelines worldwide. This discovery underscores a critical truth: Africa is a vital hub for top-tier cybersecurity talent, open-source contribution, and proactive threat intelligence. Building resilient digital economies across the continent requires continuous investment in local research and community-driven initiatives. Join the cyberstorm.mu team to participate in Project unveil today! send email to team@cyberstorm.mu. Avishai Poorun Chitrakshi Bhookhun Pushali Bundhoo Zahra Sumun Madhav Sanmukhiya #africa #ai #cybersecurity #supplychainsecurity #mauritius
Hans Balgobin replied to a comment
3d
Will Adie well this is my conservatism kicking in to find compromise I guess
Hans Balgobin replied to a comment
1w
hendrik hartmann indeed - this is very interesting and the scale of the energy savings here is on par with what I discussed with Tony. Thank you for the link!
-
Hans Balgobin liked thisHans Balgobin liked thisToday I formally take over as Head of the Electrical Engineering Division within the Cambridge University Engineering Department. As with any leadership role there is an element of bureaucracy but at the heart of the role it is about people. It is about supporting the members of the division wherever they are in their careers be they undergraduates, postgraduates, postdoctoral researchers, early-career academics, esteemed professors, technicians, administrative staff or alumni. The division is not just about world-class electrical engineering research and teaching - it also translates this into technical, industrial, entrepreneurial and societal impact. It is a community of amazing individuals who I have the honour of serving. My first term is due to last 4 years - wish me luck!
-
Hans Balgobin liked thisHans Balgobin liked thisI had the pleasure to finally present in Endocrine clinical update in stratford upon avon. I am thankful to Professor Miguel Debono for trusting me to present an interesting cushing disease case. Strong representation from team Sheffield and South yorkshire deanery... day 2 to come... with one more oral presentation from me. 🙂 https://lnkd.in/eUSrc9Nq
-
Hans Balgobin liked thisHans Balgobin liked thisMy second oral of the clinical update 2026. Really interesting interactive session. Again i am grateful to my consultants Prof Miguel Debono and Dr Allahabadia for allowing me to showcase this very rare and complex case. 🙂 I think it is a very good achievement for me personally in my final year of training. my last clinical update as a trainee. upon reflection,It has been an absolute wonderful experience in clinical update for the 3 year series.Highly recommended for trainees and early career consultants. 🙂🙂🙂 https://lnkd.in/eUSrc9Nq
-
Hans Balgobin liked thisHans Balgobin liked thisOur September edition of Commodity Insider Highlights has officially launched! We're bringing you the key session highlights from Commodity Trading Week Americas, as we get ready for our major relocation to New York in 2027. Interested in accessing expertise across the cross-commodity sector in the Americas region? Access the full edition here: https://lnkd.in/eGJUtVcW Also inside this edition… The Launch of Digitalisation in Commodities Online (10–11 November 2026) Commodity & Energy Trading Week APAC 2027 - The official agenda is live and ready to explore! Commodity Trading Week Europe 2027 - We’re unveiling our brand new Advisory Board featuring industry leaders from Glencore, Cargill, Unilever, Uniper, and more. Subscribe today to keep up to date with everything Commodity Trading Week! #CommodityTrading #CommodityTradingWeek #Singapore #London #NewYork #CommoditiesPeopleCommodity Trading Insider Highlights - The Recap You’ve Been Waiting for: Commodity Trading Week Americas’ Biggest MomentsCommodity Trading Insider Highlights - The Recap You’ve Been Waiting for: Commodity Trading Week Americas’ Biggest Moments
-
Hans Balgobin liked thisHans Balgobin liked thisGoogle researchers published a recent paper called Dream-RSI. RSI, in the context of AI, is a new term to me. In fact, I've only known RSI to stand for Repetitive Strain Injury or the VW Beetle RSi from the 2000s. In my latest video, I break down what Recursive Self Improvement means, what the paper actually shows, and why it matters for where AI is heading. Link in the comments below. #AI #RecursiveSelfImprovement #MachineLearning #GoogleDeepMind
-
Hans Balgobin liked thisHans Balgobin liked thisProject stormwing and Project unveil mentioned in defimedia.info as pan African initiatives. Using AI to discover security flaws and malware in supply chain: #mauritius #africa #ai #cybersecurity https://lnkd.in/dRwvggJGIntelligence Artificielle : faut-il vraiment avoir peur ?Intelligence Artificielle : faut-il vraiment avoir peur ?
-
Hans Balgobin liked thisHans Balgobin liked thisI spoke with iubenda about trusting AI to help us shop. Reordering coffee and booking a holiday probably shouldn’t involve the same amount of "looks good to me." 👉 https://lnkd.in/dRPg4gUB
Experience & Education
Recommendations received
1 person has recommended Hans
Join now to viewOther similar profiles
Explore more posts
Explore collaborative articles
We’re unlocking community knowledge in a new way. Experts add insights directly into each article, started with the help of AI.
Explore More