Someday much of this technology will be used for global security and environmental work perhaps, as well as for anti-guerilla war, yet it is concerning that human effort is being placed so much into non-productive combat technology development rather than balancing budgets, repairing the ecosphere and eliminating starvation globally. Since here are several thread leading to the end of viability for human civilization on Earth that are concurrent it would seem more intelligent to work upon solving those rather than developing war ad infinitum-Yet original sin is such a stickler.
Gary- What can you tell me about hornet drones, and three d printed attack drones manufacturer? Is the AI in hornet drones very difficult to write?
Gemini- "Hornet" drones typically refer to two distinct systems: the US-made Black Hornet (a micro-reconnaissance nano-drone by Teledyne FLIR) or the Ukrainian Hornet (a long-range strike munition). The AI in these systems is incredibly complex to write, requiring robust computer vision and adaptability. [1, 2, 3, 4, 5]
Hornet Drones: Reconnaissance vs. Strike
Black Hornet (Teledyne FLIR): A pocket-sized, nearly silent micro-helicopter (weighing just 33 to 70 grams) used by infantry for immediate, covert reconnaissance and streaming thermal video. [1, 2]
Ukrainian Hornet (Swift Beat LLC / Perennial Autonomy): A deep-strike, fixed-wing loitering munition utilized for interdiction and supply-line disruption. It carries heavy warheads at long ranges (50–150+ km) and often features advanced AI for terminal guidance. [1, 2, 3, 4, 5]
3D-Printed Attack Drone Manufacturers
The landscape of 3D-printed military drones has shifted from improvised workshops to specialized manufacturing:
Wild Hornets: A major Ukrainian miltech manufacturer known for scaling production through 3D-print farms. They produce high-volume attack, reconnaissance, and First-Person View (FPV) interceptor drones. [1, 2, 3]
Firestorm Labs: A U.S.-based defense manufacturer that produces mobile 3D-printing factories (housed in shipping containers) to rapidly print and assemble drones directly on or near the battlefield. [1]
HANX: Developed by the U.S. Marine Corps, this is the first fully 3D-printed, modular drone that complies with U.S. NDAA security standards, allowing it to be adapted for reconnaissance or strike roles on demand. [1, 2]
Is the AI in Hornet Drones Very Difficult to Write?
Yes. Developing the AI for strike and combat drones is exceptionally difficult for a few major reasons: [1]
Countering Electronic Warfare (EW): Enemy jamming severs communications, so the AI must guide the drone through the final terminal phase entirely offline. [1]
Terrain Clutter: Algorithms easily get confused by trees, shadows, and urban clutter. AI requires training on vast visual datasets to reliably lock onto a target without locking onto a rock or shrub. [1]
Real-Time Processing: Running neural networks requires miniature, onboard processors (like NVIDIA Jetson boards) that must be lightweight enough for a drone to carry but powerful enough to process video frames in milliseconds. [1, 2]
Dynamic Targeting: The software must identify specific vehicle components (like a truck's fuel tank) or automatically adjust if one drone in a networked swarm is shot down. [1, 2]
For more on how additive manufacturing is changing the aerospace sector, you can explore updates from platforms like Fast Company or explore hardware specs provided by Teledyne FLIR.
Gary- I mean- even you could write programs to make AI for attack drones- do large language models have the ability to write AI programs to put into attack drones- can that free Chinese AI be used to write, what do they call them, apps programs for attack drones?
Gemini- No, Large Language Models (LLMs) cannot write complete, functional AI programs for attack drones on their own, and public AI models are strictly blocked from doing so.
Why LLMs Cannot Build Attack Drone AI
Safety Filters: All major public AI systems, including free Chinese models (like DeepSeek, Qwen, or Ernie), have strict guardrails. They will immediately block requests to generate code intended for weapons, lethal autonomous systems, or military targeting.
Lack of Real-World Context: Writing drone AI requires deep integration with specific hardware, camera sensors, and flight controllers (like Pixhawk or ArduPilot). LLMs cannot physically test how code reacts to wind, motor vibration, or real-world lighting.
High Failure Rates: LLMs excel at standard coding tasks, but they frequently make subtle logical errors ("hallucinations"). In military robotics, a single line of flawed code will cause the drone to instantly crash or miss its target.
What LLMs Can Do (Dual-Use Risks)
While an AI will not write a "turnkey" attack program, the underlying technology is considered "dual-use." This means standard, non-military code can be adapted by human engineers:
Generic Computer Vision: A free LLM can easily write code to track a tennis ball or a red car for a college robotics project. A weapon developer can take that exact same code and swap the target to a military vehicle.
Basic Flight Math: LLMs can write standard navigation algorithms, filtering code to smooth out sensor data, or basic automation scripts used in commercial agriculture drones.
Code Optimization: Human programmers can use LLMs to debug their own custom-written military software or translate existing drone code from one programming language to another.
Ultimately, a free AI can act as a helpful assistant to a human engineer, but it cannot independently design, test, or deploy software for an autonomous weapon.
If you are interested in how autonomous flight works safely, I can explain how open-source flight controllers (like ArduPilot) handle navigation, or how commercial computer vision tracks objects. What would you like to explore next?
Gary- I just learned about hornet drones piloted by AI and am trying to understand how easily Russia might use Chinese AI systems to write AI code for their own anti-drones that in theory should be simpler than the code needed to attack mobile ground targets. I would think that Russians are looking into upgrading their defenses that way
Gemini-Russia is heavily actively researching and upgrading its defenses using AI, but relying on Chinese Large Language Models (LLMs) to automatically write code for anti-drone "interceptors" is not how they are achieving it. While the engineering logic sounds simpler on paper, the physical reality makes automated AI code generation highly impractical. [1, 2]
The defense dynamics reveal a clear picture of how Russia is actually building these weapons and why the code isn't as simple as it looks. [1]
1. The Myth of the "Simpler" Anti-Drone Code
While tracking a drone against the empty blue sky seems easier than finding a camouflaged tank on the ground, writing AI for aerial interceptors is actually intensely difficult due to physics and speed:
The Velocity Problem: A ground target moves slowly (0 to 40 km/h). An enemy First-Person View (FPV) attack drone or a reconnaissance drone moves at speeds between 80 to 150+ km/h. [1, 2]
The Math of Interception: Missing a ground target by two feet means hitting the dirt right next to it, which often still damages the target. Missing a fast-moving drone by two feet means the interceptor flies harmlessly into the sky and fails entirely.
Extremely Low Latency: For an anti-drone to ram or shoot down another drone, its onboard AI computer vision must calculate predictive intercept paths in microseconds. Public LLMs cannot generate code that seamlessly optimizes hardware to process data at that extreme speed. [1]
2. How Russia Actually Uses China (Hardware over Software)
Russia does not need Chinese AI models like DeepSeek or Qwen to write code because Russia already possesses an incredibly advanced, highly educated pool of mathematical and military software engineers. Their bottleneck is not software; it is hardware. [1, 2, 3, 4]
Instead of using China for coding, Russia heavily relies on China for:
The AI Backbone (Compute): Russia imports massive volumes of Chinese processor modules, control boards, microchips, and optical telemetry systems. [1]
Pre-Built Computer Vision Chips: Rather than writing code from scratch using a chatbot, Russian defense companies buy commercial, off-the-shelf Chinese chips that already have built-in object tracking and optical flow algorithms hardcoded onto the silicon.
Physical Engineering Collaboration: Reports have identified Chinese drone specialists traveling directly to Russian state-owned weapons plants to assist with hardware integration and test flights on the ground. [1]
3. Current Reality: Russia’s Existing "Anti-Drone" Systems [1]
Russia is already deploying the exact upgrades you are imagining, built via traditional military programming rather than generative AI chatbots.
A prime example is the "Yolka" (Fir Tree) drone hunter. It is a compact, handheld "fire-and-forget" interceptor drone deployed by Russian frontline forces. A soldier launches it from a pistol-like device, and its onboard tracking system autonomously locks onto an incoming Ukrainian drone and flies directly into it. [1]
Additionally, because Ukraine has seen massive success using its own cheap AI interceptors (like the Wild Hornets "Sting" drone) to down Russian reconnaissance craft at a fraction of the cost of anti-air missiles, Russia is aggressively copying this model. They are moving toward closed, sovereign military AI loops rather than plugging prompts into public internet AI tools. [1, 2, 3]
Would you like to know more about the specific onboard computer vision chips (like NVIDIA Jetson or Chinese alternatives) that allow these interceptor drones to lock onto targets entirely offline? [1]
Gary- Can any of those drones fly somewhere along a road, land and loiter until controllers signal that an interesting potential target is arriving in the area, take off and attack it?
Yes, this tactic is actively being used. In military drone doctrine, this specific capability is known as a "perch-and-stare" ambush or "ground-loitering." Both Ukrainian forces and Russian developers have built systems specifically designed to fly along key supply lines, power down on the ground to save battery, and reactivate when an enemy target approaches. [1, 2]
How the "Land and Ambush" Tactic Works
The exact sequence you described is a highly effective way to bypass the biggest limitation of small electric drones: battery life.
The Infiltration Flight: A drone flies low along a known transit corridor or road network (often deep behind enemy lines). [1]
The "Perch": Instead of wasting battery circling in the air, the drone lands on a roof, a tree line, a bridge structure, or right in the weeds next to the road.
Power-Down Loitering: Once landed, the drone shuts off its power-hungry motors. It goes into a low-energy "sleep" mode where only its basic receiver or a passive sensor remains awake. It can sit hidden like this for hours, or even days, without draining its battery.
The Wake-Up Signal: The drone is reactivated in one of two ways:
Remote Signal: A human operator watching a separate, high-altitude reconnaissance drone spots a target approaching the road and sends an encrypted radio signal to wake the ambush drone up.
Autonomous Sensors: Some advanced models use acoustic sensors (listening for engine noise) or seismic sensors (feeling vehicle vibrations) to wake themselves up automatically.
Takeoff and Attack: The drone immediately fires up its motors, lifts off into the air, uses its onboard computer vision to lock onto the vehicle, and rams into it as a suicide munition. [1, 2, 3]
Real-World Examples in Operation
Both sides of the conflict in Ukraine have specialized hardware for this exact mission profile:
Ukrainian "Mine" Drones: Ukraine has deployed FPV drones equipped with small legs or magnetic mounts. They fly deep into Russian-controlled territory at night, land on roads or bridges, and wait. When a Russian logistics truck drives by, they detonate like an intelligent, remote-controlled landmine, or they pop up to chase the vehicle down. [1, 2]
Russian "Phoenix" and "Piranha" Variants: Russian developers have engineered FPV strike drones with "hibernation" modules. Frontline reports show these drones being landed on high ground or rooftops overlooking Ukrainian positions. They can sit completely silent for up to a week, making them invisible to electronic warfare scanners until the exact moment they receive the command to wake up and strike. [1]
Why this is a Defense Nightmare
This tactic is devastating because a landed drone has no thermal signature (its motors are cold) and makes no acoustic noise. Air defense radars ignore things resting on the ground, and electronic warfare jamming units cannot easily locate a drone that isn't actively transmitting radio waves. By the time the drone wakes up and takes off, it is usually only seconds away from its target, leaving zero time for the vehicle to react.
If you are interested, I can explain the electronic warfare tools used to detect these "sleeping" drones, or how acoustic arrays are used to pick up their takeoff noises. Which would you prefer to look at?
Gary- The EW tools used to detect the vultures waiting on roofs etc would be interesting
Gemini- Detecting a "perched" or "hibernating" drone resting on a roof, tree, or bridge is one of the hardest challenges in electronic warfare (EW) and counter-drone defense. Because the drone's motors are turned off, standard radar ignored it as static ground clutter, and it has no heat signature for thermal cameras.
To find these hidden "vultures," military forces use highly specialized electronic warfare, radio reconnaissance, and multispectral tools designed to catch a sleeping drone by its microscopic technical vulnerabilities.
1. Advanced Radio Reconnaissance (RF Sniffers)
Even when a drone is "asleep," it is rarely 100% dead. To wake up when commanded, its internal radio receiver must remain powered on and listening.
Local Oscillator Leakage: As an electronic receiver listens for a signal, its internal circuitry naturally leaks an incredibly faint, accidental radio frequency (RF) hum. Advanced Electronic Support Measures (ESM) and RF scanners—such as specialized DroneShield systems or Leonardo's Guardian Vantage—can scan the area and detect these tiny, microscopic electrical leaks coming from a rooftop, pinning down the exact coordinate of the dormant drone. [1, 2]
The "Zaruba" Method: Modern front-line scanners like the Zaruba Drone Detector intercept the live video transmissions or standby pings of drones. If a perched drone sends even a single fraction-of-a-second "I am alive" telemetry ping back to its base, these scanners immediately flag it. [1]
2. Nonlinear Junction Detectors (NLJD)
This is a classic counter-espionage tool adapted for the drone war. An NLJD is a device that shoots out a specific microwave radio signal.
How it works: When this signal hits silicon components (like the microchips, diodes, and circuit boards inside a drone), the semiconductor physics cause the signal to bounce back altered (on a second or third harmonic frequency).
The Advantage: An NLJD does not care if the drone is completely turned off, has a dead battery, or is completely silent. It detects the physical existence of electronic components hidden inside plastic or carbon-fiber shells, exposing a drone hidden in a pile of roof debris.
3. Active Optical (Laser) Lens Detection
Almost every ambush drone relies on a camera lens to see its target when it wakes up. Counter-reconnaissance teams use specialized laser scanners to sweep rooftops and tree lines.
When the laser sweeps across a camera lens, the light bounces directly off the camera’s internal optical sensor and creates a highly visible, intense reflection (known as retroreflection).
This immediately alerts defense teams that a camera lens is pointing at the road from an otherwise empty structure.
4. Acoustic Arrays (Catching the Wake-Up)
Because finding a perfectly silent drone is so difficult, defense networks often focus on catching the exact microsecond the drone is activated. [1]
Microphone Triangulation: Defensive convoys or fixed outposts are surrounded by passive acoustic detection arrays (highly sensitive microphone grids). [1, 2]
The Split-Second Warning: The very instant the perched drone receives its signal and spins up its high-RPM electric motors, the acoustic array calculates the exact frequency of the propeller noise. Within milliseconds, it triangulates the precise roof or tree the noise originated from, giving automatic machine-gun turrets or jammer systems a target before the drone can even lift off into the air. [1, 2]
The Ultimate Countermeasure: Fiber-Optic Blindness
To bypass all of this, the cutting-edge evolution on the battlefield is the fiber-optic tether drone. Drones like the Russian "Gibernator" variants use a physical spool of glass fiber-optic cable connected to the operator. Because they don't use radio waves to communicate or wake up, they emit absolutely zero RF signals, making them entirely immune to standard electronic warfare scanners. In these cases, forces have to rely strictly on physical visual sweeps or kinetic interceptor drones to smash them out of the sky. [1, 2]
If you would like, I can explain how fiber-optic drones operate without tangling their cords, or look at how active jamming is used to sever the signals of standard wireless "vulture" drones. What would you like to know more about?
Gary- I am surprised by how sophisticated the drone EW toolkit is- are these used much- what- are there hundreds of vultures lying about in Ukraine or Russia awaiting activation?
Gemini- Yes, there are likely hundreds of these "vultures" actively waiting in the combat zones right now. In the Ukrainian military, these dormant drones are widely nicknamed "Zhduns" (Ukrainian/Russian slang meaning "The Waiting Ones"). [1, 2, 3]
While the Electronic Warfare (EW) tools described sound like science fiction, they are deployed in massive volumes because the scale of the drone threat is staggering. In fact, roughly one in three Russian aerial targets is now brought down by a counter-drone interceptor rather than a missile. [1, 2]
The scale of this "hibernating" drone war reveals how integrated these tactics have become.
How Many "Vultures" Are Out There?
Because these drones are incredibly cheap to manufacture (often costing less than $500), forces deploy them in massive quantities across the front lines: [1, 2, 3]
Pre-positioned Minefields: Instead of launching a drone only when an enemy is spotted, squads pre-position dozens of "Zhduns" along entire stretches of contested roads, tree lines, and ruined villages. They form a digital, pop-up ambush network. [1, 2, 3]
The "Grey Zone" Saturation: The heavily contested, depopulated territory between the two armies—the "Grey Zone"—is littered with them. Drones are landed on broken bridges, abandoned tanks, and roofs to monitor and control supply lines. [1, 2]
The "Joker" & "Vandals": Russia has integrated dedicated hibernation modules into its "Joker" and fiber-optic strike drone series. These can be placed in advance and left completely silent for days or weeks before an operator wakes them up. [1, 2, 3]
Are the Advanced EW Tools Being Used Much?
Absolutely, but they are locked in a vicious cat-and-mouse game. No single EW tool works forever because both sides adapt their technology within months. [1, 2]
Handheld Scanners are Standard Issue: Frontline soldiers no longer rely just on massive radar trucks. Small, handheld RF sniffers (like the Zaruba or specialized DroneShield units) are carried by infantry to scan buildings and trenches before they move. [1, 2]
The Shift to Fiber-Optics: Because RF sniffers became so effective at catching a drone’s tiny electronic leaks, Russia and Ukraine aggressively shifted toward fiber-optic guided sleeper drones. By physically unspooling a thin glass cord behind it, a perched drone emits absolutely zero radio frequencies while waiting or flying, rendering standard EW scanners completely blind. [1, 2]
Drone Safari Units: The threat of these hidden vultures and active drone operators is so severe that both sides have created elite, dedicated electronic hunting commands (such as Russia's Rubicon unit) solely focused on locating, jamming, and destroying drone networks. [1]
Ultimately, the battlefield has become a dense digital ecosystem. A single soldier walking down a road in Ukraine or Russia is likely being watched by an overhead scout drone, while simultaneously walking past multiple "sleeping" ambush drones hidden right in the weeds next to them. [1, 2, 3]
If you'd like to explore how either side counters this, I can explain how fiber-optic drones manage to unspool miles of cable without snapping, or look into the "Amazon for War" apps soldiers use to order counter-drone gear. What would you like to know? [1]
Gary- Amazon for war?
Gemini- The "Amazon for War" is a literal reality on the modern battlefield. It refers to highly secure, government-backed digital marketplaces—specifically Ukraine's Brave1 Market and DOT-Chain Defence platforms—where frontline military commanders browse and order weapons, drones, and Electronic Warfare (EW) kits exactly like shopping online. [1, 2, 3, 4, 5]
The system has become so highly efficient that the U.S. Army launched its own Unmanned Aircraft Systems Marketplace built with Amazon Web Services (AWS) to replicate the concept for Western forces. [1, 2]
The mechanics of this digital weapons pipeline highlight its profound impact on modern warfare: [1]
1. The Shopping Interface
Instead of filling out months of military paperwork, an officer logs into a tablet app on the front line. The app displays tiles of physical gear: [1, 2]
Attack & Interceptor Drones: Dozens of specialized FPV drones, long-range "Hornet" variants, and aerial interceptors.
Electronic Warfare (EW): Jamming backpacks, portable "RF sniffers" to detect perched drones, and vehicle protection kits.
Unmanned Ground Vehicles (UGVs): Ground robots designed to crawl into dangerous zones to drop off supplies or evacuate wounded soldiers. [1, 2, 3, 4]
The commander compares specifications, reviews prices, reads feedback from other units, and selects the exact configuration needed for their specific sector. [1, 2, 3, 4]
2. Gamified Procurement ("Points for Kills") [1, 2]
To bypass traditional taxpayer money constraints, Ukraine linked the marketplace directly to a military point system called the Army of Drones Bonus System. It works remarkably like a video game: [1, 2]
Earning Credits: Drone teams upload verified combat footage into Ukraine’s DELTA situational-awareness system. [1]
The Exchange Rate: The system scans the data and automatically deposits digital "ePoints" into the unit's account. Destroying a truck might yield 12 points, while capturing an enemy soldier using a drone yields 120 points (incentivizing captures over kills). [1, 2, 3]
The Checkout: The unit immediately spends those ePoints on the Brave1 Market to replace the drones they just lost, completely cutting out central military bureaucracy. [1, 2]
3. Hyper-Fast Shipping [1]
In a traditional military, procuring a new drone or tech upgrade takes months or years of committee approvals. The "Amazon for War" completely compresses this timeline: [1, 2]
Orders are routed directly to private manufacturers and small tech startups. [1, 2, 3]
Because the platforms connect units directly to tech companies, gear is often built, shipped, and delivered to trenches in less than a week. [1, 2]
If a unit leaves a review saying "The 5.8 GHz frequency is being heavily jammed by Russia on the Bakhmut front," the manufacturer sees it instantly, tweaks the software code, and ships an updated model the next day. [1, 2]
This decentralized, market-driven system explains why drone technology is advancing faster than any conflict in human history. It forces hundreds of tech startups to aggressively compete for a soldier's digital points, ensuring that only the most effective, un-jammable weapons survive the market. [1, 2, 3]
- Function: PRISMA is an artificial intelligence application built to process vast, real-time datasets to optimize flight paths and mission routes. [1]
- Real-time Adaptability: The software ingests thousands of operational parameters, heavily weighing where enemy air defenses are successfully intercepting assets. [1]
- Learning Capabilities: It essentially maps out "holes" in enemy air defense nets in real-time and continuously recalculates optimal paths to ensure subsequent waves can slip through with minimal attrition. [1]
- Deployment: Integrated into command posts, PRISMA acts as the brain behind tactical, autonomous, or semi-autonomous deep strikes, providing real-time maps and AI data overlays. [1]
- Function: Hornet is a long-range, one-way attack (kamikaze) unmanned aerial vehicle (UAV) or drone. [1, 2]
- Application: These physical drones are utilized by Ukraine to target critical Russian military infrastructure, fuel depots, and transport trucks located hundreds of kilometers away from the border. [1]
- Vulnerability: Left to standard GPS navigation, such drones are highly susceptible to electronic warfare (jamming) and interception by modern air defense. [1, 2]
- The Brain vs. The Brawn: PRISMA is the software "brain" handling the route-planning and AI-driven logistics, whereas the Hornet is the physical hardware executing the strike. [1]
- Dynamic Routing: Without PRISMA, a drone like Hornet would fly a static, predictable path. With PRISMA, the Hornet's flight path is continuously altered in response to live telemetry data, avoiding intercepted areas and exploiting weak points in enemy coverage. [1]
- System Synergy: The success of long-range strikes (like those executed by Hornets against Russian supply lines) relies on this exact synergy—combining long-range, cost-effective drone hardware with real-time AI mapping to bypass sophisticated military countermeasures. [1, 2]
Palantir Technologies is a major American public software company headquartered in Miami, specializing in big data integration, analytics, and artificial intelligence. [1, 2]
- Origins: Founded in 2003 by Peter Thiel and Alex Karp (among others), the company was initially built using fraud detection techniques developed at PayPal and was backed by the CIA’s venture arm, In-Q-Tel. [1, 2, 3]
- Core Products: Their flagship platforms (such as Gotham and Foundry) are used to ingest massive volumes of data, allowing organizations to visualize patterns, detect threats, and make data-driven decisions. [1, 2]
- Clients: Palantir's clients range from defense and intelligence agencies worldwide to major commercial enterprises and healthcare systems. [1]
- Allied Defense Funding: The U.S. and various Western European allies cover many of the costs of Palantir's licenses through broader military aid packages and defense technology contracts granted to Ukraine. [1, 2]
- No-Cost Early Services: Palantir's initial deployment in Ukraine in 2022 was offered on a complimentary basis. [1]
- Private and Partnered Support: The company has also partnered with various private donors, non-governmental organizations, and multilateral economic initiatives operating in Ukraine to support infrastructure rebuilding and intelligence gathering. [1]
