About
I'm a C-suite technology leader with deep expertise in generative AI, machine learning…
Activity
6K followers
-
Dr Troy Neilson shared thisToday’s office was a cracker!!! The Sydney Opera House is such an iconic and special Australian landmark. Again, great to be on the stage talking about some really interesting work with our friends at Mastercard. It’s really challenging sticking to an embargo with sooo much exciting news to share. Only a few more sleeps…! #ai #agenticai #agentpay #agent Glassbox Labs
-
Dr Troy Neilson shared thisWhat a day, what an amazing start to the week, and what a way to place a cherry on top of months of really hard work..! Auckland really turned it on! Sunshine and beautiful weather. Couldn’t have asked for more. As always, the team at Mastercard were the perfect hosts, and have been an absolute pleasure to work with. A special shout out to Raj Seshadri, Paul Monnington, Surin Fernando, Ali Hawli and Gaurav Bansal, with an extra special shout out to my mate Praveen Kumar as well, who helped guide us with unwavering patience through these uncharterred waters, and helped us navigate the world of Mastercard - Rock Stars all of them!! Off to Sydney at the Opera House on Wednesday, then to Melbourne on Friday to cap off the week. It’s killing me I can’t say more… yet! Follow me, Glassbox Labs and Mastercard for more “globally” epic and exciting news to come!!! I’ll keep saying it, what a time to be in this space… #ai #agenticai #agentic #agent #agenticcommerce #agentpay #sme #smeai #llm #gpt #auckland #nz
-
Dr Troy Neilson shared thisAnthropic’s new Opus 4.7 seems to be really good. I’ll be looking at what improvements can be seen in the coding domain, as they seem to be leaning heavily on this in the documentation. If you’re harnessing around it, perhaps go have a look at the new prompt guides to make sure you’re employing best practices. Not sure about the “model welfare” side of things..! Happy to discuss this. Exciting times… #claude #opus #ai #llm #aiharness #aiagent #agentic Glassbox Labs
-
Dr Troy Neilson shared thisWe've been crazy busy recently with all things agentic AI. Looking forward to some exciting announcements starting next week... #agenticai #ai #agenticpayments #agent
-
Dr Troy Neilson shared thisSooo.... If, like me, you are: * An AI nerd; * An Apple nerd (fanboy, etc...); and, * An NVIDIA nerd, Then this is REALLY interesting. For the first time since 2019, you can now run NVIDIA GPUs natively on Mac thanks to the team at the tiny corp. This is early days, and the performance is meh, but as we see everyday in this space, things move quickly, so it's likely performance will jump inline with uptake, so fingers crossed... This is amazing work, and it brings a smile to the dial knowing that smart teams like these guys are out there and motivated to go after these really interesting challenges. Again, what a time to be in the AI game..! #ai #gpu #llm #macOS #Mac #nerd #tensorcore #blackwell #rubin #hopper https://lnkd.in/gdRjgky5
-
Dr Troy Neilson shared thisProject Glasswing looks like the moment cyber defence became an AI race Anthropic just made one of the most consequential announcements in AI safety to date. Project Glasswing unites Amazon Web Services (AWS), Apple, Cisco, CrowdStrike, Google, JPMorganChase, The Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks around a single mission: use frontier AI to find and fix critical software vulnerabilities before attackers do. Their unreleased Claude Mythos Preview model has already autonomously discovered thousands of zero-days, including a 27-year-old flaw in OpenBSD and a bug in FFmpeg that survived five million automated tests. It looks like we’ve crossed a threshold where AI outperforms all but the most elite human security researchers at finding and exploiting vulnerabilities. That capability isn’t going back in the box. Anthropic is backing Glasswing with US$100M in credits and $4M to open-source security organisations. At Glassbox Labs, this resonates deeply. The architecture required to autonomously reason about codebases and chain exploits is fundamentally an agentic systems problem, the same class of problems we’re building for. Where agentic cyber capability goes today, every other complex domain follows tomorrow. For Australia, this should be a wake-up call. Our critical infrastructure runs on the same software stacks Mythos Preview just punched holes through. The Glasswing coalition is exclusively US-anchored big tech. The question for Australian policymakers: where do we fit in AI-driven cyber defence? The glasswing butterfly hides in plain sight. So do the vulnerabilities in the software that runs our world. The era of AI-augmented cyber defence just arrived. I keep saying it, but what a time to be in this game..! 🔗 https://lnkd.in/gYe__5iA #AI #Cybersecurity #ProjectGlasswing #AgenticAI #SovereignAI #Anthropic #GlassboxLabs #australia #claude #mythos
-
Dr Troy Neilson reposted thisDr Troy Neilson reposted thisA single GPU failure in a 1,024-GPU cluster costs over $300,000 per month in lost compute. Most teams are still checkpointing every few hours. When one node fails during distributed training, the entire job stops. The other 1,023 GPUs go idle while the system detects the failure, replaces the node, restarts the job, and loads the last checkpoint. Everything computed since that checkpoint is gone. The math is concrete. At $3/GPU/hour with a 1,024-GPU cluster, a 7.9-hour mean time to failure, 1.5-hour checkpoint frequency, and 15 minutes of recovery overhead, the wasted compute adds up to $307,000. That number comes from real production data, not a theoretical model. MTTF decreases as cluster size increases. A failure that happens once a week on a 128-GPU cluster can happen multiple times per day at 1,024 GPUs. The checkpoint frequency that worked at smaller scale becomes ruinously expensive at larger scale. This is the single biggest reason GPU infrastructure ROI falls short of what finance teams modeled. The hardware works. The GPUs are fast. But every failure rolls back hours of progress and burns money on idle compute that produces nothing. Checkpoint strategy is not an engineering detail. It is a financial decision. How often does your training pipeline checkpoint, and have you calculated what a single rollback actually costs?
-
Dr Troy Neilson shared thisImagine running a massive ~300B parameter MoE-LLM on an off-the-shelf MacBook Pro… This very thing has been achieved by running a really large MoE model with Pure C and Metal. The real clincher is that this isn’t something that dropped from one of the big labs, this came from one person. Some truly amazing and inspired work by Dan Woods, VP of AI and ML at CVS. I keep saying it, but what a time to be in the AI game..! #ai #purec #metal #llm #gpt https://lnkd.in/eFcW7MPAGitHub - danveloper/flash-moe: Running a big model on a small laptopGitHub - danveloper/flash-moe: Running a big model on a small laptop
-
Dr Troy Neilson shared thisWhat Simon Villani, PhD said!Dr Troy Neilson shared thisWhat a dramatic day. —————————- I’m currently building Villani Code, a local-first coding agent that beats Claude Code on small-model benchmarks. GitHub: https://lnkd.in/gxUUHAvB
Dr Troy Neilson commented on a post
2d
Fantastic write up Barbara Cresti
-
Dr Troy Neilson liked thisDr Troy Neilson liked this📄 It's my data and local open source, 'off-grid' models can be useful. I paired OpenClaw with Google's Gemma 4 31B running locally and it genuinely adds value. Why bother investing time and effort? 1️⃣ - Token costs and subscriptions will continue to rise. I wanted something without limits, without a meter running, and without someone else's terms of service deciding what I can build. 2️⃣ - Corporations will need people who can set up and manage "off grid" value. Running your own agents, on your own hardware, with your own governance. That skill set barely exists yet. 3️⃣ - When you own the model, nobody deprecates it on a Tuesday (we're looking at you recently Anthropic). No surprise API changes. No vendor swapping the engine mid flight. A stable foundation you can build around, experiment with, and break with minimal consequence. #AI #TechStrategy #OpenSource Blog and write up in the comments.
-
Dr Troy Neilson reacted on thisDr Troy Neilson reacted on thisI always knew the Kiwi could fly!!!! Super proud to see Westpac NZ shine at Microsoft AI Tour today! Well done to Danielle, Ravind, and the entire WNZL team! #MicrosoftAITour
-
Dr Troy Neilson reacted on thisDr Troy Neilson reacted on thisIt becomes official once the sign is up!!! Our new SHARON AI office is thriving in North Sydney. Allowing us to be closer to partners within our ecosystem. What I love the most is walking in each day to a full house of super smart, positive, energetic teams all working towards a common end goal! Lets go team #SHAZ!!! James Pickens
-
Dr Troy Neilson reacted on thisDr Troy Neilson reacted on thisMy team is growing! I'm looking for a Managing Consultant to strengthen the team- delivering projects for one or more of our clients. The initial role would be as senior product owner (debit cards) for one of our Sydney FI partners. If that's you- please apply! https://lnkd.in/gnJpGgYnManaging Consultant, Advisors & Consulting Services, DeployManaging Consultant, Advisors & Consulting Services, Deploy
-
Dr Troy Neilson liked thisDr Troy Neilson liked thisMost people credit us for our trusty cards, but can you name all the ways we support global commerce? From powering a global network that connects millions of merchants and financial institutions, to helping detect fraud in real time, to enabling seamless identity verification and AI-driven experiences — Mastercard is part of the infrastructure that keeps the digital economy moving. It shows up in everyday moments: 🚇 Tapping into transit 🌍 Sending money across borders 🔐 Checking out without entering card details ⚠️ Getting alerted to suspicious activity Behind those moments is a focus on making payments smarter, safer and more accessible — while helping businesses grow and economies become more inclusive. Swipe through to see how we show up beyond the card. Learn more: https://lnkd.in/etZUfPDj
-
Dr Troy Neilson liked thisDr Troy Neilson liked thisIt's begun! Work on NEXTDC's GE1 4.4MW Data Centre in Geelong has commenced bringing new capability to the region. With a DA approval of just 28 days, I'm excited to share the cracking pace we deploy the critical digital infrastructure to Corio. It takes a village, but the safety standards and systems Kapitol bring to the site build are close to NEXTDC's heart. They allow speed, quality and efficiency whilst still ensuring that everyone goes home safely and in the same shape. Thanks for site tour with the crew onsite including Jamie, Sam, Chri, Evelyn, Sud, Luke and George from Newbridge, and Imogen from WT (Australia). Not long now until the electrons will be flowing. #whereaithrives #wherethecloudlives #digitaleconomy
-
Dr Troy Neilson reacted on thisDr Troy Neilson reacted on thisA billion francs. Ten years. 54,000 workstations. Switzerland is walking away from Microsoft. The announcement lands weeks after Microsoft 365 finished rolling out across the federal administration. Bern deployed the system, saw it working, and still concluded the dependency was unacceptable. The objection is structural: the US CLOUD Act. Passed in 2018, it gives American authorities the legal power to compel US companies to hand over data - regardless of where that data physically sits. A server in Zurich offers no protection. A Swiss data residency clause offers no protection. The request can arrive at Microsoft, be executed globally, and the end user may never be notified. ➡️ For Switzerland, whose diplomatic posture rests on the confidentiality of what crosses its borders, that is a sovereignty issue. ▫️ Germany saw it first. Schleswig-Holstein is migrating 30,000 public sector workers to LibreOffice and Nextcloud, with annual savings of €15M. ▫️ France has been building DOCAPOSTE and OVHcloud into sovereign stacks for years. ▫️ The European Commission is funding Euro-Office for a 2026 launch. ▫️ Switzerland's transition will be slow, expensive, and politically contested. That is why it matters. What boards & execs should pause on Microsoft's sovereign cloud, Google's EU data boundary, AWS's European Sovereign Cloud - they are technical responses to a legal problem. They harden the walls. They do not change the CLOUD Act. ➡️ Data location and data jurisdiction are different questions. The consequence: vendor concentration is being reclassified For over a decade, running everything on a single hyperscaler was considered operational discipline. Now it reads as governance risk - by auditors, regulators, investors, and customers asking where their data is legally exposed. StratEdge was built to surface a risk that boards are only beginning to price in: AI and cloud are not separate decisions. Every enterprise AI deployment inherits the jurisdictional exposure of the infrastructure it runs on - the training data, the inference logs, the prompts. Most EU enterprises are spending between 1.5% and 3% of revenue on cloud and SaaS infrastructure with meaningful CLOUD Act exposure. For a mid-size company, that is often eight figures annually, locked into a single jurisdiction, with no modelled exit cost. Switzerland paid a billion francs to arrive at the question every board should be asking now: what is the defensible number, and what is the real one? #DigitalSovereignty #Boardroom #AIGovernance #StratEdge
Experience & Education
-
Glassbox Labs
View Dr Troy’s full experience
See their title, tenure and more.
or
Licenses & Certifications
Publications
-
Deals of a feather… Modelling latent classes in R&D collaboration data using finite mixture analysis
PLOS ONE
This work explores if behaviour-based asymmetries are likely to impact deal valuation in the life sciences by examining positive public sentiment as a proxy for market behaviour when negotiating under asymmetric conditions to examine heterogeneity in research & development collaboration (RDC) deal data. We use public sentiment as a proxy for behaviour along with stage of development-based RDC deal data to search for latent classes in the deal data using finite mixture modelling. The analysis…
This work explores if behaviour-based asymmetries are likely to impact deal valuation in the life sciences by examining positive public sentiment as a proxy for market behaviour when negotiating under asymmetric conditions to examine heterogeneity in research & development collaboration (RDC) deal data. We use public sentiment as a proxy for behaviour along with stage of development-based RDC deal data to search for latent classes in the deal data using finite mixture modelling. The analysis reveals a nuanced picture: public sentiment emerges as a significant predictor of deal value, but only for approximately 15% of the data set. This subset exclusively includes firms in the Preclinical stage, where projects have moved past discovery but are yet to commence human studies. Interestingly, the research finds that sentiment’s impact on deal valuation is particularly pronounced in this stage, suggesting heightened market sensitivity. With recent research demonstrating that knowledge asymmetry and behaviour impact valuation volatility, we take this further by capturing latent classes within the data which demonstrates how behaviour is most influential in deal pricing considerations. We argue that our research demonstrates the impact of asymmetry and market behaviour on a subset of RDCs where products are known, but likelihood of success is difficult to determine.
Other authors -
Organizations
-
Institute of Managers and Leaders
Fellow
-Fellow of the Institute of Managers and Leaders (IML) ANZ
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