From Petri Dishes to Quantum Proteins

Gemini Generated Image

I’ve written here a bit about my undergrad degree in microbiology, The path to that degree came with a strong dose of biochemistry along the way. UMass Amherst, mid-1970s. I worked through enzyme kinetics, the Michaelis-Menten equation, and the standard biochem textbook of the day, Lehninger's Biochemistry. Then I spent a few years in a hospital clinical lab, plating cultures and squinting at gram stains, before I shifted into electrical engineering. The proteins we now talk about in quantum computing papers were, back then, things you drew mechanisms for on an exam and tried not to confuse.

When I saw the May 19 Phys.org piece on a 12,635-atom protein simulation, the old biochem brain lit up. Cleveland Clinic, RIKEN, and IBM used a quantum-centric supercomputing workflow to model two real proteins, T4-Lysozyme and Trypsin, sitting in water with the small molecules they bind to. Trypsin is a serine protease from your pancreas. It cuts other proteins at specific spots, after the amino acids lysine and arginine. T4-Lysozyme is an enzyme that breaks down peptidoglycan, the mesh that holds bacterial cell walls together. Both are fundamental biochem topics.

So.... what does "simulate a protein" mean? A protein is a folded chain of atoms. It does its job because of its shape and because of how the electrons inside it are arranged. Simulating it means using physics to figure out where every electron is and how it is moving. That information is what tells you whether a drug molecule will fit into the protein and stick, or just bounce off. The bigger the protein, the harder that calculation gets, and the difficulty grows fast.

Six months earlier, the same team did a 303-atom Trp-cage, a tiny model protein biochemists use as a folding test case. This run jumped to 12,635 atoms. That is a 40-fold increase in size and a 210-fold gain in accuracy on a key step. The hardware was IBM's 156-qubit Heron r2 quantum processors at Cleveland Clinic and RIKEN, working alongside Fugaku and Miyabi-G, two of the largest classical supercomputers in Japan. The team ran the quantum machines for more than 100 hours and collected 1.3 billion measurements.

So why is quantum better here? Think about it this way. Electrons in a molecule don't sit still like marbles. They behave like little waves that overlap and influence each other. To keep track of just one electron on a regular computer, you need to store some numbers describing its wave. Add a second electron that interacts with the first, and you don't double the work, you square it. Add a third, you cube it. The work doubles with every electron you add. Ten electrons is easy. Fifty electrons fills up a big supercomputer. A hundred electrons would need more memory than there are atoms in the Earth. A protein has thousands.

A quantum computer is built out of the same kind of stuff the electrons are made of. Each qubit is itself a tiny quantum object that can hold the wave behavior directly. So instead of storing a giant list of numbers to fake the physics, the qubits just do the physics. Fifty electrons need about fifty qubits, not a supercomputer. One hundred electrons need about one hundred qubits. The problem stops blowing up. Richard Feynman pointed this out in 1981: if you want to simulate nature, use a computer made of the same stuff as nature.

Back to classical computers for a moment. There is a method that gets the right answer exactly, called full configuration interaction. It is the gold standard. The problem is that every electron you add explodes the work, which is exactly the blowup I just described. It works fine for a water molecule. It is impossible for a protein.

So computational chemists fake it. They use an approximation called density functional theory, or DFT, which gives a fast, decent answer for most of a molecule. DFT is the workhorse of computational biochemistry today. It has a known weak spot, though. It struggles in the active site of an enzyme, the small pocket where the actual chemistry happens and where a drug would bind. That is the one place you need the answer to be right, and it is the one place DFT gets shaky.

That is where the hybrid setup comes in. Use the quantum computer on the small, tricky pocket where DFT fails. Use the classical supercomputer on the rest of the protein and the surrounding water, where DFT works fine. Each machine does the part it is good at. Neither could do this protein alone. Together they can.

Hybrid classical-quantum is how real chemistry gets done in the near term, not pure quantum and not pure classical. Biochemistry is one of the applications pulling the field forward, especially structure-based drug design. And atom count is becoming the benchmark people compare against, the way transistor count once was for chips.

Beaming Power From a Cessna

Back in 2023 I remember reading this writeup on Caltech's Space Solar Power Demonstrator. They beamed a small amount of microwave power between panels on a satellite in LEO. Neat physics, but I have heard the space-based solar pitch for thirty years and the engineering never closes. Launch mass, spectrum rights, pointing accuracy, regulatory approval for a multi-megawatt beam over populated areas. Something in that list always kills it.

Then I read this Spectrum article about Overview Energy. They flew a Cessna turboprop over Pennsylvania at 5,000 meters in 70-knot crosswinds and held a power beam on a ground receiver the whole time. Watts, not kilowatts, but the first time anyone has done it from a moving platform at altitude. They beamed near-infrared light, not microwaves. That choice is the part worth looking at.

Microwaves have two problems for ground power delivery. The good bands between 2 and 20 GHz are already spoken for. 5G, GPS, satellite links, weather radar, military. You will not get a license to transmit megawatts in any of them. The second problem is beam spread. Diffraction limits how tight you can focus, and at microwave wavelengths the spot on the ground is kilometers across even from a big aperture. That means a rectenna farm the size of a small city.

Optical wavelengths change the math. Same aperture, beam spot drops from kilometers to meters at GEO range. No spectrum fight either, since power transmission at optical frequencies is not allocated the way RF is. The receiver question gets interesting too. Overview wants to drop the IR onto existing utility-scale PV farms. Silicon is not tuned for a single IR line so you lose efficiency at the receiver, but you also skip building a new class of ground infrastructure and skip the permitting fight that comes with it. Whether the efficiency hit is worth the deployment savings depends on numbers we have not seen yet.

Now the scaling, which is what every engineer reading this already knows is the hard part. A working demo at the bench is one thing. A working demo a thousand or a million times bigger is a different machine, with different failure modes, different cost curves, and different second-order effects that did not exist at small scale. Heat dissipation that is trivial in a benchtop laser becomes the dominant design problem in a megawatt source. Vibration modes that did not matter in a fixed lab fixture wreck pointing accuracy on a 100-meter deployed structure. Manufacturing tolerances that were acceptable on one unit are not survivable across the thousands of components a flight article needs. Cost per watt that pencils at the prototype scale almost never holds when you push three or four orders of magnitude. This is the part of engineering that nobody outside the field appreciates and everyone inside it has been burned by.

Apply that to Overview. The Cessna run was watts. DARPA's July 2025 demo pushed 800 watts across 8.6 km for half a minute. Overview's roadmap calls for megawatts from GEO by 2030 and gigawatts later. Five to nine orders of magnitude. Continuous operation instead of a 30-second pulse. Pointing accuracy in the microradian range from 36,000 km up, on a satellite that heats and cools every orbit and flexes accordingly. The Cessna proves they can track a target from a moving platform. GEO needs the same trick done a thousand times better, with thermal management for a source running at a million times the optical power.

The spacecraft itself is the other hard part. To collect useful sunlight you need a big aperture. A big aperture does not fit in a rocket fairing, so it has to fold for launch and deploy in orbit. JWST showed that can work. JWST also showed how close that kind of mission comes to dying. Add debris survivability over a 20-year design life, station-keeping fuel budgets, and disposal at end of life, and you have a spacecraft program as risky as the beam.

So where does that leave me. Still skeptical, but less than I was. The 2023 Caltech work was a physics check. The Cessna flight is a systems check. Source, beam control, pointing, tracking, and a PV-style receiver, all running together on a moving platform in real weather. If Overview gets a LEO demonstrator up and lands a kilowatt on the ground from a folded-then-deployed aperture, people will be paying close attention.

What Jensen Huang Thinks Smart Looks Like Now

Years back I had a student who was terrible at exams. Multiple choice, timed problems, placement exams, all of it. He barely squeaked by in lecture. In lab he was someone else. He would stand next to a partner who was building a circuit and just watch. Then he would point at a cap and ask what happens if it fails open. He was usually right before the thing ever got powered up.

He got a field tech job after he graduated. Two years in, customers were calling and asking for him by name. He still cannot tell you the formula for a low-pass filter off the top of his head. He can tell you, on the phone, that the problem you think you have is not the one you actually have.

I thought about him this week reading about Jensen Huang on Jodi Shelton’s A Bit Personal Podcast. She asked him who the smartest person he has ever met is. He would not answer. He said the question itself is the wrong one now.

Huang’s point: the kind of smart we have always rewarded, the technical problem solver, is becoming a commodity. He used software programming as the example. For decades that was the smart-person job. Now it is the first thing AI is doing well.

So what does he think smart looks like going forward? Three things stacked. Technical understanding, human empathy, and the ability to pick up on what nobody is saying. That last one he called seeing around corners. It comes from data, analysis, first principles, life experience, wisdom, and reading the people in the room. People who have that mix tend to spot problems before they happen.

This is part of why I started doing oral exams in my Circuits class. A written test will tell me a student got the right number. It will not tell me whether they know why they used mesh instead of nodal, what a time constant actually means, or where the sign error came from. Sit them down and make them talk through it and you find out fast. 

Most of what engineering school grades on is the part AI is taking over. Timed problem sets. Multiple choice. Take-home work that gets judged on how polished it looks. The stuff Huang is talking about, reading a situation, asking the question nobody else thought to ask, knowing something is off before the numbers say so, is harder to grade. It is also what is going to matter.

My old student would have bombed every test Huang is calling outdated. He read a lab bench the way some people read a chessboard. Twenty years ago that made him a good tech. Today it makes him the kind of engineer you cannot swap out for a chatbot.

From $14 Billion to $125 Million: What Killed Chegg

I had a Chegg account for years. For my electrical engineering classes, I would check my  own homework solutions using Chegg before handing them to students, just to double check my work. It was a good tool and worth the money at the time.

Then one night I tried Google Gemini on the same kind of problem. It gave me the answer, walked through the steps, and did a better job explaining what was going on. I sat there for a minute and thought, why am I paying for Chegg? I cancelled the next morning. These days Gemini is what I lean on for problem solving in my classes.

Turns out a lot of people made the same call. Chegg was worth around $14 billion in 2021. Today it is worth about $125 million. The company laid off almost half its remaining staff last October. Revenue last quarter dropped 42% from the year before. The CEO points the finger at ChatGPT and at Google's new AI answers that show up right at the top of the search page. Students get their answer before they ever click through to a paywall.

Khan Academy is a different story but feels some of the same heat. It is a nonprofit, so there is no stock price to watch. A few years back they rolled out Khanmigo, an AI tutor, and got it in front of more than 1.4 million students. On paper, big win. In April, Sal Khan admitted most students barely touched it. The reason is not hard to figure out. Free AI is already open in another tab. Kids are not going to switch over to a separate tutor when help is right there.

Both companies were built to sit between a student and an answer. AI now sits closer to the student than either of them. When that happens, the middleman business gets thin in a hurry.

The Oral Exam Experiment Worked

Last fall I posted that I was dropping homework from the grade book and adding an oral portion to every exam in the spring. Students were running Engineering homework problems through Gemini and handing in solutions they could not explain when I asked. I was grading a chatbot, so I stopped.

Spring semester is over. It worked. The oral portion runs about ten minutes per student. They pick one problem from their written work and walk me through it. Why mesh and not nodal. What the time constant tells you about the circuit. Where the negative sign came from. I learn more about what a student actually understands in that ten minutes than I used to learn from a semester of graded homework.

A new Lumina Foundation-Gallup study says 57% of US college students use AI in their coursework at least weekly, and one in five use it every day. At the same time, 53% say their school discourages or prohibits it. Daily use is highest among men and among business, tech, and engineering students. The students avoiding AI mostly cite ethical concerns and school policy, so the ones following the rules are falling behind on a tool they will use the rest of their careers.

My position on AI in education is simple. If we are preparing students for the jobs they are about to take, AI has to be in every class. Every engineering job they walk into will expect them to use these tools well. A program that prohibits AI is training students for a job market that no longer exists. The work is not to keep AI out of the classroom. The work is to teach students how to use it, where it fails, and when to check it against first principles.

That still leaves the assessment problem. If students use AI on everything, how do you know what they understand? You change how you measure them. Oral exams catch what written work cannot. In-class paper problems catch it. Hands-on labs, where a student wires a circuit on a breadboard, takes scope measurements, and explains what they are seeing, catch it cold. Take-home essays graded on polish do not catch anything anymore.

The AI can solve the circuit. It cannot explain why this student chose the loop they chose, and it cannot wire the breadboard when the lab is due at five. That is what we should be assessing, and that is the work employers are hiring for.

Car Buyer vs User

Image Google Gemini Generated

I stopped at a dealership recently to get some parts for my wife’s car and, as always, took a peek in the showroom. I ended up looking at stickers and counting the screens in a mid-range SUV. Four. One in the dash, one in the middle, one for the passenger, one in the back. A salesperson grabbed me and started walking me through them. I stopped paying attention about 30 seconds in. Paddle shifters, massage seats, gester controls, ten drive modes, ambient lighting, self parking, some kind of fragrance thing. I nodded. I would use none of it.

Apparently I am not alone. The JD Power 2024 Tech Experience Index rates the so-called advanced driver features near the bottom of what owners actually value. Gesture controls hit 43 problems per 100 vehicles, which is a polite way of saying they do not work. People pay for this stuff, give up on it, and then never bring it up again until they trade the car in.

I keep thinking about the iPhone in this context. Two billion of them have shipped. None came with a manual. Kids figure them out before they can read. Apple kept taking things away. The home button. The headphone jack. Each cut made the phone easier. Carmakers do the opposite. They keep piling features on and call it progress.

I drive a 2020 Tesla Model 3. The software is great. The maps work, the updates show up overnight, the menus make sense. The car also has a turn signal stalk, a gear stalk, and two scroll wheels on the wheel. Tesla removed all of those in the 2024 Highland refresh. You signal now by pressing a button on the spoke. You shift gears by swiping a slider on the screen. Owners are paying around $400 for aftermarket kits to put the stalks back in. Tesla rolled out a version in China with the stalk returned. The best software in the industry could not save the worst ergonomic decision in the industry.

A few years ago BMW went the other way and tried to charge for hardware. Eighteen dollars a month to turn on the heated seats already wired into the car. About 90 percent of BMWs leave the factory with the hardware in place. Customers were essentially being charged twice. BMW killed it in September 2023. New Jersey introduced a bill to make hardware paywalls illegal on cars, which tells you how badly that went over.

None of this is accidental. Cars get designed for the showroom because that is where the sale happens. Three knobs and two screens looks cheap next to haptic glass and 47 menu layers. Marketing wants the acronym count on the window sticker. So you walk in as a buyer, feel reassured by the complexity, sign, and drive home. About ten minutes later you become the user. The user wants the heat on without looking at a screen, the mirrors adjusted without a settings tree, and the left turn signaled with one finger (not that one! )

Things are starting to turn. Mazda kept the knobs and gets credit for it every time someone reviews the car. Volkswagen pulled the buttons, watched sales slip, and admitted in 2025 the buttons are coming back. Their design chief said the reversal is permanent. Euro NCAP is moving toward requiring physical controls on five core functions to keep a five-star rating, which means the regulators are about to do what the marketing departments would not.

The buyer is in the showroom for an afternoon. The best interface is one you do not have to think about, and you should not need a manual to signal a left turn.

DC, AC and GROUND Coupling Settings on the Oscilloscope

A student stopped me in lab last week. She was staring at a trace that kept drifting off screen. She had the coupling set to DC and a large offset was pushing her signal out of range. The question she asked was a good one: what is the difference between AC, DC, and ground coupling, and when do you use each?

The short answer is that coupling controls what the oscilloscope passes to its vertical amplifier before it draws anything on screen. With newer scopes from different manufacturers it is sometimes difficult to find where to set coupling so you can either hunt around (usually the way I do it!) or consult the user manuals.The choice changes what you see, and getting it wrong means you either miss something or start chasing a problem that is not there. Here's a pic with short definitions for each following.

DC Coupling

DC coupling passes everything through: the full signal including any DC offset. If you have a 10 mV sine wave sitting on top of a 12V supply, you see both. This is the correct default for most measurements. Power supply work, logic signals, anything where the offset carries information requires DC coupling. You want the complete picture. That student had the scope set to DC Coupling (often the default.)

AC Coupling

AC coupling inserts a series capacitor that blocks DC and attenuates frequencies below roughly 10 to 20 Hz, depending on the scope. That same sine wave now appears centered on zero, and you can use full vertical scale to examine it. This is useful for noise on a supply output, audio signals, or serial data eye diagrams where the DC level is irrelevant. The tradeoff is real: the capacitor charges and discharges, so the trace drifts for a few seconds after you switch modes. Square waves also suffer at low frequencies because the flat tops droop as the capacitor charges through the input impedance.

Ground Coupling

Ground coupling disconnects the input entirely and connects the vertical amplifier to ground. The signal disappears. You get a flat line at 0V. This is a calibration step, not a measurement mode. Use it to find your ground reference position before you connect a live signal, especially when you need to compare multiple channels on the same scale.

The student switched to AC coupling, centered the trace, and saw exactly what she wanted to see: about 80 mV of sine wave.