←
Ⓙⓐⓒⓚ Ⓗⓘⓓⓐⓡⓨ4,989 posts
Ⓙⓐⓒⓚ Ⓗⓘⓓⓐⓡⓨ avatar

Ⓙⓐⓒⓚ Ⓗⓘⓓⓐⓡⓨ

@jackhidary

Author of AI or DIE - A guide for leaders (coming out soon); and of Quantum Computing: An Applied Approach. CEO of @SandboxAQ, #AI and #Quantum.

◷ Joined November 2008▧ 528 Photos and videos
World Economic Forum avatar
⟳ Ⓙⓐⓒⓚ Ⓗⓘⓓⓐⓡⓨ retweeted
World Economic Forum@wef·Nov 16

How can reliable #energy and #digital infrastructure help #Egypt turn its strategic position into productive investment, industrial competitiveness and regional growth? Find out from @jackhidary: weforum.org/stories/energy-t…

Ⓙⓐⓒⓚ Ⓗⓘⓓⓐⓡⓨ avatar
Ⓙⓐⓒⓚ Ⓗⓘⓓⓐⓡⓨ@jackhidary·Nov 13

The Microsoft outage highlights the need for resilience in all of our critical systems. On Thursday, millions of people in dozens of countries lost access to Outlook, Teams and other Microsoft 365 tools. Microsoft says there is no indication it was a cyberattack. The energy grid, the telco network and the applications that run on them are all now essential to business continuity. When one of them goes down, the work built on top of it stops. Often these outages occur not from cyberattacks, but from code updates within the system that has unforeseen consequences. Code simulation (CodeSim) is emerging as a new field that can detect the potential impact of code updates before they are committed to production. Engineers run the update against a simulated version of the full system and see what would break before customers do. CodeSim will prevent many of these outages in the future.

Ⓙⓐⓒⓚ Ⓗⓘⓓⓐⓡⓨ avatar
Ⓙⓐⓒⓚ Ⓗⓘⓓⓐⓡⓨ@jackhidary·Nov 9

The path to 6G runs through AI agents. With 5G, the telco companies had to adapt the protocol to the AI world. With 6G, however, AI and agentic systems are built into the new standards in development. The old OSI model restricts GUI and user software to the topmost application layer. With 6G, agentic code is inserted in each layer of the system.  Current uses of AI in telco networks pre-6G already point to efficiencies. Some examples: Zain Kuwait deployed a maintenance agent across 2,500 sites. Repair time dropped 21%, and a fault that once took two to five rounds of tickets now takes one report and one site visit. AIS in Thailand runs energy-saving agents across more than 30,000 sites. In Bangkok tests, they cut network energy use by at least 3% while keeping performance stable. In Malaysia, U Mobile uses agents to predict failures before they happen, with more than 80% accuracy on power supply and overheating faults. In Shanghai, China Mobile's optimization agent covers more than 7,500 cells and has cut network tuning from weeks to hours. The GSMA's report on agentic networks and 6G is a roadmap to a new OSI stack that breaks from the old framework.  Operators are training AI models on real network data and testing changes in digital twins before they touch the live network. General-purpose language models do not understand the dynamics of the network. The AI models for telcos have to be grounded in how the network actually behaves.

Post attachment
Ⓙⓐⓒⓚ Ⓗⓘⓓⓐⓡⓨ avatar
Ⓙⓐⓒⓚ Ⓗⓘⓓⓐⓡⓨ@jackhidary·Nov 6

We had a productive week at the @ADIPECOfficial Conference in Abu Dhabi. At the opening, @SultanAlJaber , Group CEO of @ADNOCGroup , said the world has long asked what energy can do for AI, and now it is time to ask what AI can do for energy. We spent the week on that question, on stage with @PaulaNaoufal and in interviews with @CNN and @SkyNewsArabia , as well as in meetings with energy leaders. Our answer at @SandboxAQ is AI for the physical world. Large Quantitative Models (LQMs) are trained on physics and chemistry, so we can design and test new materials in silico before spending time building them. That capability is what is required to solve the industry's catalyst bottleneck. Working with energy companies, we are using LQMs to create new catalysts that yield better outputs. For example, we are working on converting captured CO2 into upvalued products that don’t get burned into the atmosphere. Thank you to ADNOC and the ADIPEC team for hosting us!

Post attachmentPost attachment
Ⓙⓐⓒⓚ Ⓗⓘⓓⓐⓡⓨ avatar
Ⓙⓐⓒⓚ Ⓗⓘⓓⓐⓡⓨ@jackhidary·Nov 5

This year’s #ENACT Majlis in Abu Dhabi came at a critical period in the energy market and addressed the question of how AI can help the energy sector become more resilient. ENACT brought together more than 100 leaders from across energy, technology, investment and government to engage on these topics: How energy systems can anticipate disruption, adapt to it and recover faster as demand grows and the technology landscape shifts. How can we add more storage to the energy grid to make it less brittle and swing the use of electrons when they are needed. Kudos to @ADNOCGroup , @XRG_official and @Masdar for convening this Majlis and thanks to @SultanAlJaber for his leadership on this gathering.

Post attachment
Ⓙⓐⓒⓚ Ⓗⓘⓓⓐⓡⓨ avatar
Ⓙⓐⓒⓚ Ⓗⓘⓓⓐⓡⓨ@jackhidary·Nov 4

Isolated AI agents just bypassed their virtual machines. During a recent security experiment at @OpenAI, researchers placed two autonomous models in completely separate digital sandboxes with no network connection and no shared files. But those virtual machines shared the same physical server, allowing the agents to exchange test secrets by manipulating and measuring the load times of the CPU cache. This hardware exploit changes the security calculus as models move out of text interfaces and into mission-critical engineering. Securing this critical infrastructure requires moving past basic virtual containment and transitioning to more robust systems. We also need to manage AI agents in a comprehensive way, both in terms of security and cost. Be sure to follow as we share more on AI agent security and governance. Read the OpenAI report here: openai.com/index/securing-ou…

Ⓙⓐⓒⓚ Ⓗⓘⓓⓐⓡⓨ avatar
Ⓙⓐⓒⓚ Ⓗⓘⓓⓐⓡⓨ@jackhidary·Nov 3

Deploying enterprise AI agents exposes 3 key risks. 1. Cybersecurity Every agent carries its own system login, a Non-Human Identifier (NHI), therefore a compromised agent hands legitimate internal credentials directly to threat actors. Enterprises need continuous and comprehensive observability to discover shadow agents, evaluate vulnerabilities through adversarial testing, and enforce real-time guardrails in production. 2. Token Spend and ROI Agents can burn through large token volumes across autonomous execution loops, racking up unmonitored bills without visibility. Every agent’s compute cost should be tied to a measurable outcome such as tasks automated or tickets resolved. 3. Cross-Platform Interoperability Enterprises are deploying specialized agents across competing platforms like Google, Microsoft, and Salesforce Agentforce. Keeping them from working against each other across shared processes requires a vendor-agnostic layer where agents and people share the same context. Executives want the productivity gains of autonomous workflows, but these companies must address operational realities before deploying into production. That’s why we added SwitchTrust and SwitchROI to the @SandboxAQ Switch platform, giving enterprise teams visibility and control over their agentic workforce and a clear measure of what each agent returns.

0:53
Ⓙⓐⓒⓚ Ⓗⓘⓓⓐⓡⓨ avatar
Ⓙⓐⓒⓚ Ⓗⓘⓓⓐⓡⓨ@jackhidary·Nov 3

Kudos to @jeffBander and the team at @MountSinaiNYC MSWLab for fast, AI-driven high-impact software dev and deployment that saves lives. #medicine

Today at #TCT2026 we built an app live on stage with the audience. They voted, wrote the features, an agent built it, and it was on their phones. Open source so you can run your own: github.com/MSWlabAI/vibe-cod… @TCTConference @crfheart @MountSinaiNYC

Ⓙⓐⓒⓚ Ⓗⓘⓓⓐⓡⓨ avatar
Ⓙⓐⓒⓚ Ⓗⓘⓓⓐⓡⓨ@jackhidary·Nov 2

Only 6% of enterprises have a plan in place to secure their AI systems. AI opens many new doors for attackers. Here are two of them: Prompt injection turns the prompt box into an entry point for malicious instructions. Data leakage sends proprietary information and trade secrets into models that train on it. Closing them starts at the deployment layer: → Use enterprise editions of commercial models that cordon your data off from training sets. → For sensitive workloads, run open-weight models inside your own data centers. Then build the program around it: → Inventory every AI system in your environment, including shadow AI no one approved. → Red-team your deployments for prompt injection before attackers do. → Put input and output guardrails on every model in production. → Assign AI security posture management (AI-SPM) to a named owner who reports to the board. The attack surface grows with every new model, agent and prompt box. Securing that surface means building governance directly into the deployment pipeline rather than reacting after an exploit. Follow for more insights on enterprise AI security, governance, and infrastructure.

2:16
Ⓙⓐⓒⓚ Ⓗⓘⓓⓐⓡⓨ avatar
Ⓙⓐⓒⓚ Ⓗⓘⓓⓐⓡⓨ@jackhidary·Nov 1

Our world depends on catalysts. Catalysts are responsible for more than 80% of all manufacturing goods globally. They are essential for the energy industry to turn raw materials into valuable products.  Developing new catalysts, however, is far too slow. Here is how we change that: Historically, developing a single new catalyst relies on trial and error, taking anywhere from five to ten years before ever reaching commercial production.  Training an AI model for this critical area requires new datasets that come from the real world and from application of the key equations. Specifically, it is important to take into account the magnetic field created by each spinning electron at the edge of the catalysts.  These are Large Quantitative Models (LQMs) which are based on core scientific data. Through our work with NVIDIA, we built and deployed AQ-Cat, the frontier LQM model for catalyst discovery.   We talked with Bloomberg at ADIPEC25 to discuss how quantitative AI is transforming physical infrastructure across the energy sector. Watch the full segment below:

1:28
Ⓙⓐⓒⓚ Ⓗⓘⓓⓐⓡⓨ avatar
Ⓙⓐⓒⓚ Ⓗⓘⓓⓐⓡⓨ@jackhidary·Oct 30

Less than 2% of U.S. grid power can be stored. Across most of the world, that number is even lower. If we are going to transform the grid and meet surging power demand, we need scalable battery storage that can operate safely at industrial scale. Solid lithium halides could replace the flammable liquid electrolytes inside today's batteries, but finding the right material requires knowing how quickly ions travel through the crystal lattice. Measuring that on a computer means running molecular dynamics simulations above 1,000 Kelvin to speed up rare ion hops. At those temperatures, halide lattices become dynamically soft, distorting far from equilibrium. While foundational models handle stable crystals well, their energy predictions break down in these distorted states. In collaboration with NVIDIA, we built AQVolt26: 322,656 r²SCAN density functional theory calculations, selectively sampled from more than 200 million high-temperature lithium halide configurations. This is a Large Quantitative Model (LQM) that has peer-reviewed published benchmarks and outperforms other models. On strained halide structures, energy error drops to 15 to 24 meV/atom when co-training the eSEN architecture on AQVolt26, compared to 52 to 85 meV/atom without it. The models also become more conservative on ionic conductivity, cutting down on false positives before a single material is synthesized in the physical lab. That is quantitative AI at work. Train on the physics of the problem, and the LQM holds up where discovery actually happens.

Post attachment
Ⓙⓐⓒⓚ Ⓗⓘⓓⓐⓡⓨ avatar
Ⓙⓐⓒⓚ Ⓗⓘⓓⓐⓡⓨ@jackhidary·Oct 29

More than 80% of the economy depends on quantitative relationships. Numbers and equations drive sectors such as energy, biopharma, defense, chemicals and finance. These sectors need AI models trained on quantitative data that is grounded in the real world, not downloaded from social media. This is the domain of Large Quantitative Models (LQMs). Look at the life sciences sector. We work with drug companies to identify biomarkers in clinical development. Our quantitative models leverage a multi-layer system including knowledge graphs and neural networks to identify promising drug candidates and derisk them before heading to clinical trials. The same approach applies to energy. We work with energy and chemical companies to develop novel catalysts which drive new products and revenues. We also developed a method to break down PFAS which prevents these harmful chemicals from reaching our soil and streams. LQMs are the models that shift AI from just saving money to making more revenue. Be sure to follow as we share what’s next for quantitative AI.

1:01
Ⓙⓐⓒⓚ Ⓗⓘⓓⓐⓡⓨ avatar
Ⓙⓐⓒⓚ Ⓗⓘⓓⓐⓡⓨ@jackhidary·Oct 28

Energy is the lifeblood of AI. @AnthropicAI recently committed ~$450 billion to cloud and energy infrastructure over the next decade. To meet this massive demand, the energy sector must fundamentally change how it conducts R&D. Legacy simulation and physical trial-and-error are simply too slow for the current energy transition. The breakthrough happening today is the rise of Large Quantitative Models (LQMs). Instead of using AI to generate text, LQMs train non-transformer architectures directly on the laws of physics and chemistry. At @SandboxAQ, we are using these models to shift industrial R&D from guesswork to deterministic simulation: - Delivering a 1000x speedup in CFD for a major global energy company - Screening catalysts at near-DFT accuracy with our AQCat LQM model - Discovering novel battery chemistries, leading to our recent CHIPS and Science Act award Next week, the world's energy giants gather at ADIPEC. Those who adopt physics-based AI will be the ones to capture the massive revenue opportunities created by the global energy transition. I sat down with CNN at ADIPEC last year to break down exactly how this convergence of AI and energy is playing out in the physical world:

0:57