Dragon1-1

You should do training missions before say things like that, because you're quite wrong. The ones that are limited don't take months (and don't break on every update, either), and the ones that do take months are very clever and flexible. The only thing AI could do better there is, perhaps, recording the voice lines. DCS scripting sucks, that's another problem. The real reason that mission development can take months is that tools we get are difficult to use. No AI will ever be able to bypass inherent brittleness of DCS scripts, simply because it can't be bypassed without a scripting overhaul on the engine side. ED seems to be trying to make ME less painful to use, but it's a slow process. AI is not magic, nor is it a person. In the end, it's just a bunch of equations and a lot of brute force. As a tool, this can be useful, but it won't do the thinking for you. Those "things only human can spot" only seem hard to guess, but that doesn't necessarily mean they are. You can use it to rephrase existing information in a way that suits you better, but be careful not to assume too much about what it can do.

This is exactly what would happen if someone were to invent a new fuel, or a better engine. Moreover, this example is particularly bad, because rockets are designed to run on a specific type of fuel. Sure, people would be going to Mars in fewer launches, but not on SpaceX rockets. That is why they'd take a stock hit, and also why other AI companies are taking the hit now. It's not their AIs that are in demand. Likewise on the semiconductor market, if it can work with fewer, lower end NVIDIA chips, who's to guarantee the next version won't run on, say, a few 2070 from the bargain bin? Stock market bets on promises and expectations, and right now, it took a hit because there's a very reasonable fear that the AI market will not continue its current explosive growth trajectory, but will instead plateau, with companies moving towards efficiency instead. Except the consumers want to go to Mars, not to a giant ball of gas shrouded in radiation. Maybe, for instance, parts for Mars bases do not get cheaper or faster to produce just because rockets did, and bases designed on Mars will not work on Jupiter. In that case, even if SpaceX reengineers their rockets to work on the new fuel (itself a big investment that will eat into profits), the result will be that fewer rockets will fly through no fault of their own. Understand that, and you'll understand why companies sometimes choose a FUD campaign instead of getting on with the times. Sometimes, it's enough to outright kill off an innovation. Another example: if you build a better mousetrap in a mice-infested country, you'll certainly make good money. But what if it's so good that it outright extirpates the mice? Would you be able to sell an even better trap in this situation? I don't think so. In some markets, it's very much possible to go out of business because your product was too good. For further reading on why your theory is bunk, look up Phoebus cartel. Yes, which is why most of of what you wrote is neoliberal balderdash. Your thinking is exactly what made the markets, and the world as it is now, such a mess. There is no such thing as infinite demand, the only reason NVIDIA can say things that it does is that they're a near-monopoly when it comes to AI chips, and thus are free to manipulate the supply. It's not that demand is infinite, it's that the people making the stuff made damn sure to never make enough chips for everyone, even though they could. Markets are a complex thing. They're never completely free, they depend on finite resources, finite logistics, finite market size, and are often interconnected (for instance, a shortage of latest socket CPUs will bog down sales of mobos with that socket, no matter how good they are). Real markets, as opposed to simplified models used to bamboozle the general public, are never as well behaved as you'd like to think. I'd suggest you educate yourself on this before you pretend to know anything about economics (and I really wish politicians heeded that advice...).

Note that the only thing you need an AI in this scenario is natural language processing, in other words, talking to you like a human. The rest is, in essence, a glorified flight manual and basic flying handbook reader, cued up by cockpit actions. Effectively the same thing DCS training missions already do, autogenerated in real time. I admit, Deep Seek makes a pretty good job at the talking part. Seeing as it's lighter on the hardware than other models, it could be a good starting point.

You can set them so that the key does not command full deflection of the stick. However, a realistic sim will be unplayable without analog controls almost by definition.

That was my point. My dad was very much in the race for a long time, in fact, he'd been at it since he swapped his second Atari for a PC. As a kid, hand-me-downs from him were sufficiently powerful that I seldom complained about not being able to run something, except when things like SSE2 appeared, at the time I had the last high end AMD CPU that didn't support it, which kind of sucked when it became a hard requirement. At one point, he just decided further upgrades weren't worth the money, the 20xx series RTX being a bit of a dud had a lot to do with it, and the 30 series launch pricing sealed the deal. The "target consumers" are getting tired of ballooning prices for incremental performance gains. While I haven't yet declared the 3090 will be the last GPU I buy, unless something truly revolutionary is added (more than spamming fake frames, anyway), I expect it to suffice for the foreseeable future. When I do upgrade my rig, it'll likely be because something like retina-level VR came out.

My dad actually did. It's been a while since he got himself a 1080-based rig. Nearly perfectly silent when idle or close to idle, can run almost everything but the very latest stuff on his modestly sized screen, and unlike my own rig, it doesn't double as a space heater. He barely uses its full capabilities, anyway. In fact, I suspect that quite a few people dropped out of the performance race, which is causing a lot of grief when they try playing the latest games on a rig that was cool when it was built (like my dad's), but the technological progress passed them by. In fact, I'd expect the majority of people to not be too eager to constantly update their PCs. I think it's way past time for that. UE5 games, in particular, seem to be suffering from an epidemic of bad optimization and poor coding that can be powered through, but people shouldn't have to buy a top tier GPU just to play the latest releases. Even lowering the graphics settings often doesn't help.

Whatever the beancounters think makes them the most money. Idle them, store the output in a warehouse, or retask them. I don't know what they'll do, but lowering prices is unlikely to be part of the plan.

What NVIDIA is saying can also be said in other words: "pay and cry, we know full well we haven't squeezed every possible dollar out of you yet, and we'll just keep throttling supply if you try any of that do more with less nonsense". With Trump tariffs, expect prices to only get worse. TMSC will likely not come to the US, since that would mean paying US taxes and US wages. It could be cheaper for the company to leave its customers paying tariffs, and even if they do build a US factory, chips from that will be expensive. Samsung does make chips, too, by the way, in addition to Chinese manufacturers who are slowly catching up, so it's not that all our eggs are in one basket, but either way, the market is not very competitive, and any hiccup from TMSC will result in others jacking up their prices. This "business" is all about the reason why the pricing on the 50 series is as insane as it is, specifically how NVIDIA is blatantly abusing the mechanism. The other discussions are not that far off topic, either, since they all circle back to two fundamental questions about the 50 series: "why can't I afford the one I want?" and "do I need that overpriced crap at all?".

Here's a thought: someone should make a modular GPU board. Instead of selling everything in one pack, let us buy a board, the GPU chip itself, however much VRAM we care to insert, and a cooler for all that. It works that way with the rest of the PC, why not the GPU? I don't think inserting a CPU-style socket would affect performance too badly.

Or, you can sell at a loss, and sell a product or service that will make up for that loss (of course, for this you do have to make the "killer app" which will be both indispensable and unusable without your hardware). Several examples of that from the tech world, as well. Besides, mass production itself tends to bring costs down a lot. That's what I was referring to, "cheap enough to mass produce" means figuring out how to make it work with a Chinese assembly line. I assure you, consumer electronics aren't sold anywhere near cost. Electronics are typically sold at a huge markup, unless someone is trying to pull the aforementioned trick (which is why it works in first place). Economics of scale mean that once you have mass production going, you can slash the markup. In fact, I think that laser diode arrays might one day scale better than screens do. One nice thing about retinal projection is that it doesn't necessarily need to go to the same extremes of miniaturization that screens have to in order to look good (same reason why you don't get screen door effect when you project a full HD screen to cover an entire wall). Added bulk isn't too bad, and miniaturization is what tends to drive up manufacturing costs.

They still do broadcast, though. You can fool some older ESM systems, particularly automated filters and the like (common on RWRs, which need to avoid adding too much pilot workload), but in the end, a dedicated EW platform like found in large SAM sites or on a surface ship will be able to look at raw emissions spectrum and see that white noise is stronger from a given direction, even if only slightly. The moment you start radiating anything, you risk giving yourself away. Of course, that alone doesn't necessarily mean you'll be engaged, but it does mean the enemy might think to point some more powerful radar at the area where the unusually strong white noise is coming from. LPI doesn't mean undetectable, it just tries to trick automatic filters into filtering it out.

With more emphasis on "digital" and less on "warrior" as the era goes on, too. In fact, given how much of a flying iPad the F-35 is, I suspect it's paving the way to eventually sticking the pilot permanently into the simulator and wiring the cockpit to a broadband datalink, with the jet flown remotely. And that would sneakily pave the way for the beancounters to finally fire those pesky fighter pilots and replace them all with relatively cheap, mostly replaceable drones that don't complain about flying at night, don't get busted for DUI after a night in town, and don't threaten to go work for the airlines every time you "misplace" their promotion papers to save a bit of dough on personnel expenses. The next generation of fighters is already planned to be "optionally piloted" (read: drones with a cockpit stuck in to prevent the fighter community from lynching the designers). I'm not saying it's a good idea, I'm saying someone is going to try. One hopes that by the time someone figures out how to jam or worse, hijack the datalink, there's still someone in active service who not only knows how to fly a plane, but can train others to do it.

Can't something be done with what the AH-64 uses for rocket zones? This is a similar concept, you have a pylon with one of each missile.

Good video. I wonder if you picked up on the irony of talking about simulating "mature, adult professionals", in light of the early part, when you mentioned that early debriefs would often degenerate into "No, I shot you down" back and forth. DCS community and actual fighter pilots who get to handle million dollar fighting machines have more in common that one would hope...

You don't have to add algorithms to DCS. After all, the aircraft models don't change much mid mission, and when they do, they do so in a few, predictable ways. Nobody cares how exactly RCS changes during gear transition, you can calculate RCS with gear up and gear down, then LERP between the values. Hence, the algorithm would be used to precalculate RCS tables for each aircraft (and maybe even for ground units).

It is an S duct, just a shallow one. Either way, you need to be quite a bit under the nose to see the turbine blades. Notice how low is the angle of that picture, and look at the model in DCS if you don't believe me. Also, even without RAM, radio waves are not 100% reflected, and also notice that the interior of the intake doesn't have flat surfaces. This would cause most of the waves to also bounce to the sides, and this would result in them exiting the intake at an angle, not to mention such a complex path would obscure the strobing effect of turbine blades (which is distinctive enough that a mid-2000s Viper can tell which plane it's looking at based on this). You can't really treat an intake like this as a flat surface. MiG-21 doesn't have this advantage because the fan disk on all MiGs prior to -23 is in plain view, or obscured only by the inlet cone, which is transparent to radar anyway because that's where its own radar is. That would have the effect of making the Viper more visible to SAM radars than to other fighters, because the former would have a good view of the turbine blades. In fact, if you're sneaking up on a fighter, you'd probably want to be below it, to force it to look for you against ground clutter. Aggressor Vipers are well known for their bushwacking abilities, and even against blue sky, you'd often see them late (definitely later than you'd want, given the A model was a vicious WVR combatant).

No (at least the C model and earlier, they might have added something in newer blocks), but it has an S-duct, which, thanks to its angled interior wall, means it doesn't quite count as a flat surface. To figure out how the radar waves actually come out after bouncing around inside, you'd need to use the modeling software, but my intuition is that a lot of them will be scattered, rather than reflected back. Also note that we're comparing it to other jets of the era, in the vast majority of which the fans are directly in view from the front, with fan blades perpendicular to the airflow to within a few degrees, providing a nice, bright strobe. I'd imagine that Viper is not the only jet that would give interceptors trouble in that area, but merely the most famous. I wonder how the F-104 will look with proper RCS modeling, for instance. It's also diminutive, very pointy, has a small radome, S-ducts (albeit shallower ones than on the Viper) and a limited missile load. I haven't heard too many stories about fighting this one, but I imagine that once properly modeled, it'll appear on the radar quite late, as well. The Viggen might, as well, at least when not loaded for bear.

It looks like we might be hitting physical limits of what 2D silicon chips can do. While a 3D architecture is a possible solution, I'm a bit worried about heat dissipation, particularly with multiple layers of logic.

Elon Musk: hold my beer... Besides, quality doesn't matter if you can compete on price, so yeah, you very much can mass produce something that you don't think is good enough, if you can make it sufficiently cheap. Of course, that tech is far from cheap, but there are several other points that it could compete on, like weight or weight distribution. Also, that viewfinder is not that far below what the early Oculus Rift versions could do, which was usable for VR.

IIRC, this was said about the A model, so it wouldn't be festooned with AIM-120s, rather it would only carry four AIM-9s, and probably no bags as well, or just on the centerline. We're also talking aircraft mounted radars of the 80s, so they weren't quite as good as what we have now. It'd still show up noticeably later on modern sets, especially if relatively clean, but it wouldn't be as sneaky as the A could be. That said, if a modern Block 50 Viper drops both the bags and the air to ground hardware, it could still be pretty sneaky, especially if it has just two AMRAAMs and a pair of Sidewinders.

No, just annoyingly hard to detect, according to pilots who flew against it. In absence of the turbine disc, the radar is probably the biggest contributor here. It's not a Nighthawk, but it's still small enough to show up on radar noticeably later than most other jets.

That's true, which is why the exact mechanics of the RAM don't matter, you can just assume some percentage of inbound radiation is absorbed instead of reflected no matter what angle, and this translates to a straightforward multiplication by a ballpark estimation or RAM efficiency. The vast majority of stealth comes from shape, specifically from avoiding presenting a flat surface to the radar antenna. Radar waves reflect just like any other waves, angle of incidence equals angle of reflection. The tricky bit here is hiding the engines and making an intake that will not reflect waves straight back from its inner surfaces while still working well as an intake. This is also why the Viper is a surprisingly good stealth aircraft, despite not really being designed as such: it's physically dinky and its intake has a pronounced S-duct, so the turbine fan (a big honkin' radar reflector) is not directly in view from most angles. The F-35 is different from the Viper in that it also pays attention to reducing side and rear RCS, in addition to hiding the fan disk even better.

Eyelids will get in the way no matter what (that's their job), and while the eye can change shape, shape of the pupil is directly correlated with its optical properties. Yes, it's different per person, but this only means some way of adjusting the HMD to the specific wearer is necessary. You can try tracking the retina, but it's not the only way to tackle this problem. None of them require a major breakthrough, though. It's more of a matter of making the thing affordable for mass production, this tech is still in its early phase, and that means improvements are likely to come fairly quickly. If patent issues don't choke it off, we'll probably see it in a consumer HMD before the end of the decade.

There's actually public software that can be used to determine RCS of any given shape. The only thing that's classified here is how well the RAM coating actually works, and that would just multiply the RCS from shape by a certain factor. The physics behind radar wave propagation are not classified, nor even particularly complex (though heavy to calculate in real time).

Retinal projection is already a thing, however it seems to me that it's going to be more of an AR solution, unlikely to displace screens completely. It's already a commercial technology, though, with Retissa viewfinder using it. For what it's worth, to make a VR HMD you don't actually need to track the retina (which, after all, is inside the eye), you need to track the pupil. It is an interesting proposition, since unlike with the screen, you could potentially locate the emitter (and thus most of the weight) somewhere else than in front of the eyes, and use fiber optics to deliver the beam. Fiber optics don't like bending, so it'll probably still be on your head, but it could be somewhere more comfortable.