The Enginuity Digest
Sometimes I hear whispers that the pace of AI innovation is finally slowing down. Then, a week like this happens. In the span of just a few days, we’ve watched NVIDIA post earnings that rival the GDP of small nations, global stock markets drop and bounce back in a frenzy of volatility, and Google release a state-of-the-art reasoning model alongside a massive image upgrade codenamed 'Nano Banana' (odd name, I know). Throw in the fact that AI is now learning to drive CAD software like a human, and it’s clear that if anything, AI development might actually be accelerating
Here’s what will be covered in today’s newsletter:
News Update:
Google releases Gemini 3.0: A new benchmark for reasoning and multimodal engineering.
MIT’s new AI agent learns to use CAD software just like a human.
NVIDIA obliterates earnings expectations (again) amidst global market jitters.
Deep Dive: Hybrid Models—Why the Marriage of First Principles and AI is the Future of Engineering.
AI in Application: Simulating Reality with Gemini 3.0.
In Other News: Liquid metal science, pumped hydro news and Australian industry updates.
What’s been happening with AI?
Google’s New Gemini 3.0 Breaks All Benchmarks (Link) |
Google has officially launched Gemini 3.0, its newest flagship model, and for many people, this is the one they’ve been waiting for.
Gemini 3.0 has crushed the leaderboards, notably topping almost every single benchmark (apart from a couple of coding ones). It even scored 41% on "Humanity’s Last Exam", a benchmark designed to be nearly impossible for current AI (The next best is GPT-5 Pro at 30.7%).
The model features state-of-the-art multimodal reasoning. It can ingest massive technical documents, schematics, and video feeds simultaneously and "reason" across them. Most notably is its visual reasoning which performs well beyond all other models.
The model is designed for "agentic" workflows. This means developers can build tools where the AI doesn't just give advice but takes action, querying databases, running calculations, or interfacing with other software.
Why this matters: I thought it would be hard to convince me to stop using ChatGPT, but Gemini 3.0 has completely changed this. We are moving past "chatbot" territory and moving to systems that handle complex, multimodal data that brings us closer to true automated assistance. It’s less about generating funny images (though "Nano Banana Pro" can do that too) and more about reliable, complex problem solving.
The best bit is that uni students can get a free annual pro subscription if you sign up before December 9th.
MIT’s new 2D-to-3D Model that Design’s Like an Engineer (Link) |
Researchers at MIT have released a breakthrough model dubbed "VideoCAD", which takes a completely different approach to 3D generation.
Instead of generating a 3D mesh from a text prompt (which often results in unusable blobs), this AI agent was trained on thousands of videos of humans using CAD software.
The AI takes a 2D sketch and actually operates the CAD interface—clicking buttons, drawing lines, and extruding shapes—to build the model step-by-step.
This ensures the final output is a fully parametric, editable CAD file, rather than a static mesh.
Why this matters: This is the missing link for generative engineering. Generating a pretty 3D picture is useless for manufacturing; generating a parametric file with a feature history is the holy grail. This agentic approach could automate the "drafting" phase of design entirely, something that is both fascinating and scary.
NVIDIA’s Financials vs. The Economy of Volatility (Link) |
Credit: moneycontrol
NVIDIA reported earnings this week that can only be described as insane, with revenue soaring 62% year-over-year to $57 billion, beating even the loftiest Wall Street whispers.
CEO Jensen Huang dismissed "AI bubble" fears, stating that the demand for Blackwell chips is "off the charts" as companies race to build physical AI infrastructure.
However, the broader market is less convinced. US markets saw sharp volatility this week, driven by renewed inflation fears and uncertainty over Federal Reserve rate cuts.
Why this matters: NVIDIA is the proxy for the AI industry’s health. Their continued explosive growth confirms that the infrastructure build-out (data centres, energy, cooling) is not slowing down. For engineers, this means the capital expenditure boom in energy and advanced manufacturing is secure for now, regardless of what the broader stock market does.
Hybrid Models: Combining AI with Science
We love AI for its speed, but we don't trust it with our bridges. Why? Because standard Deep Learning models (like the LLMs powering ChatGPT) are probabilistic "black boxes." They don't know Newton's Laws; they just know statistical correlations. If you ask an LLM to simulate a fluid flowing through a pipe, it might hallucinate a scenario where water flows uphill because it "looks right" based on its training data.
On the flip side, traditional engineering relies on First Principles simulation, things like Finite Element Analysis (FEA) and Computational Fluid Dynamics (CFD). These are rigorously accurate but computationally expensive with things like modelling the aerodynamics of a new turbine blade taking days on a supercomputer.
Enter Hybrid Models: The Best of Both Worlds
We are currently witnessing the rise of Physics-Informed Neural Networks (PINNs). This is a hybrid approach where the "loss function" of the neural network isn't just minimizing the error against a dataset; it is penalized for violating the laws of physics.
In simple terms: You teach the AI the equations that govern the system. If the AI predicts a solution that violates conservation of mass or energy, the model "punishes" itself during training.
The Engineering Impact
1000x Speedups: Once trained, a PINN can infer the result of a fluid simulation in milliseconds, whereas a traditional CFD solver might take hours. This allows for real-time Digital Twins. Imagine a control room operator seeing a real-time heat map of stress on a reactor vessel, updated instantly as they change pressure setpoints.
Inverse Problems: Traditional solvers are great at "Forward Problems" (Given these forces, what is the deformation?). PINNs excel at "Inverse Problems" (Given this deformation sensor reading, what were the forces?). This is a game-changer for predictive maintenance and non-destructive testing.
Sparse Data: Because the AI knows the physics, it doesn't need millions of training data points. It can "fill in the blanks" between sparse sensor readings using the laws of physics as a guide.
The Path Forward
We are moving away from AI that "guesses" and toward AI that "solves." Companies like Ansys and Siemens are already integrating these hybrid solvers into their suites. For the engineer, this means the ability to run thousands of design iterations in the time it used to take to run one. For more info check out these:
AI in Application: Simulating Reality with Gemini 3.0
Gemini 3.0 can even make quantum computer simulations
With the release of Gemini 3.0, we have reached a tipping point where "Vibe Coding" (writing software using natural language) isn't just for making websites, but now also for making engineering tools.
Because Gemini 3.0 is an "Agentic" model, it can plan, write code, execute it, and refine it in a loop. This means you can now build interactive simulations in minutes without needing to be a software developer.
Here are three actionable examples of how it could easily be implemented:
1. The Interactive Process Dashboard (HTML/JS) Imagine you want to visualize a simple tank system for a junior engineer. You can prompt Gemini:
"Build me a process dashboard. It should have a slider for 'Inflow Rate' and a slider for 'Outflow Valve Position'. Show a live animated chart of the Tank Level over time. If the level hits 90%, flash a high-level alarm."
Gemini 3.0 will write the HTML, CSS, and JavaScript code to model the differential equation of the tank level and render it in your browser. You get a functional, physics-based interactive tool in 30 seconds.
2. Discrete Event Simulation: Need to optimize a logistics queue? You don't need complex software for a quick check.
"Write a Python script to model a mine truck weighbridge. Trucks arrive every 10 minutes with a standard deviation of 2 minutes. We have 2 scales. Each weigh-in takes 5 minutes. Simulate a 24-hour day and tell me the average wait time. Then, run it again with 3 scales and compare."
The AI understands the stochastic nature of the problem, writes the discrete event simulation, runs it, and gives you the statistical output.
3. Logic & Safety Verification: You can upload a logic narrative (e.g., a Control Philosophy) and ask the AI to simulate it to find "doom loops."
"Here is the logic for our pump startup sequence. Simulate the state machine step-by-step. What happens if the 'Flow Switch' fails to activate within 5 seconds but the 'Stop' button is pressed at second 4?"
Gemini can "trace" the logic paths and identify race conditions or deadlocks that a human review might miss.
How to do it: You don't need to install Python or VS Code. You can go straight to Google AI Studio, select Gemini 3.0 Pro, and just type your request. It will generate the code and, in many cases, run it right there in the browser. This democratizes simulation, turning every engineer into a tool-builder.
In Other News
RBA Minutes Reveal Inflation Concerns: The Reserve Bank of Australia signaled that while the economy is slowing, underlying inflation remains sticky. (Link)
New "Metallic Gel" Created: Researchers have developed a pure metallic gel that is highly conductive and flexible. This could be the breakthrough needed for next-gen soft robotics and liquid metal batteries (Link).
Kidston Pumped Hydro Goes Live: For the first time in 40 years, a new pumped storage scheme has been registered in the National Electricity Market (NEM). The Kidston project (a repurposed gold mine in Queensland) is officially on the books, proving that old mines can learn new tricks (Link).
CSIRO’s Major Restructure: Australia’s national science agency announced a controversial pivot this week, cutting jobs in mineral resources to double down on the "clean energy transition." A bold move, but one that has the traditional mining sector raising eyebrows (Link).
Tomago Aluminium’s Lifeline: Negotiations are underway to save Australia’s largest aluminium smelter using power from Snowy Hydro. Here, industry meets the renewable transition, possibly needing a government parachute to succeed. (Link)
OpenAI is Worried: In a leaked memo, Sam Altman (CEO of OpenAI - who owns ChatGPT) has expressed concern over revenue changes following the new release of Gemini 3.0. (Link)
This newsletter seeks to engage and challenge the way engineers see AI and its potential for application in industry. Any thoughts, questions or arguments are welcome! Finally, if you enjoy the content, please refer it on to your friends and colleagues, I would love the support.
