A text-to-speech (TTS) model. Most good TTS models are closed-source. I intend on making this one open-source.
All the decent open-source ones are fairly basic with limited fine tuning and no alignment (RLHF).
I plan on adding those things. Although I am not sure if there will be any demand for it. Plus, there's a decent chance meta will make llama 4 speech output making this one obsolete.
I have often wondered : Why don't they attach a transponder on each athlete's center-of-chest (using some objective definition) and use that to time each athlete? That can be much much more accurate than any camera.
We can still use cameras for visual confirmation but transponders are much more accurate than any camera.
I'm not convinced they don't. If you look at Lyle's bib when he takes it off and shows it to everyone, there's something on the back. I thought it was another YuGiOh card at first, but then I thought, "That'd be the perfect place to put something to trigger some sort of check at the finish line."
I suspect it's probably just easier for them to print all the race numbers the same way, but only use the transponder on races where it doesn't need to be accurate, or they need to take several measurements throughout the race.
For the longer distances, the lap times and finish times show up almost instantly when there's a big gap between athletes, although I believe there's a laser at the finish line too.
Even if that is technically feasible that wouldn’t account for the lean, neck length etc. the center of the chest isn’t an accurate measurement given the current rules.
Can you explain what you mean by that? The only thing I can think of is that GPS has 2 modes : one civilian and other military use. But military use-case uses encryption so civilians can't use that mode anyway.
GPS was built primarily for the military and the signals available for civilian use were crippled to reduce accuracy. GPS receivers that provided high accuracy fell under the ITAR rules (ie you needed proper licensing to trade them)
Over time such restrictions have largely been lifted, but there are still US export controls on GPS receivers with particular features, such as those designed for use in high speed aircraft, those able to decode the still-military-only encrypted signals that piggyback GPS and provide greater positioning precision, or those designed for use in rockets/UAVs.
So, it's reasonable to assume that if you were to build your own, and do too good of a job, you would accidentally become subject to arms trade regulations, and that's probably not a place you want to unintentionally find yourself, particularly if you're publishing it as open source on the Internet :)
> those able to decode the still-military-only encrypted signals
How? I was under the impression that military-only signal was encrypted. And if someone breaks that encrpytion, blame should go to the poor handling of encryption rather than the person breaking it? Analogy : if you leave a classified document on the train and a passenger reads it, whose fault is it?
The Kraken RF project, which has multiple SDR radios next to each other, had demo code of a passive radar: the slight time delay by which signals arrived at the antennas could be used to detect large metal bodies (IOW: planes) in the sky.
They took down the demo code they were informed that they were violating ITAR regulations.
I'm relatively certain that the US government will be unmoved by your argument that it would be their fault if you built a GPS receiver that could decrypt the precise positioning signals ;)
Well, consider another analogy: a person leaves their door unlocked and another person goes in their home. It is still trespassing, even though the physical barrier has been intentionally disabled.
I'm not sure even what section of US code to check but it is certainly plausible that you can (for example) find a surplus device that decodes the signals on EBay and it's actually illegal to do it. You can find radios on EBay that broadcast on bands that are unlawful to use (or unlawful without certain licenses, &c).
I never really thought of that. That's a pretty interesting restriction. Although any party with access to warheads that can fly 1000+ mph probably can bypass the GPS restriction, no?
There was a guy in NZ (Bruce Simpson) who detailed on his blog how to make a DIY cruise missile. I think he got politely asked to stop doing so at some point.
BPS.Space on youtube is working on a DIY space-capable rocket and in a recent video he mentioned that he is not doing this as a tutorial and that his guidance system likely already wanders into ITAR territory, and thus he's self-censoring which parts he shares and which parts happen off-camera.
Ultimately these kinds of regulations are fairly silly because a sufficiently determined smart person can recreate the covered technologies from scratch, but here we are.
Any sufficiently smart person can accomplish the same original work as any other sufficiently smart person.
But when the details of these things are published or otherwise made openly available, it doesn't take nearly as many smarts to duplicate these accomplishments.
Quite often, that's good: It's easy for a dullard like me to build a circuit or to re-use some clever assembler code when someone else has published it for my own tinkering around the house. In this way, it's a pretty great world to live in; it is often very simple to stand on the shoulders of giants and get some things done that I could probably never do on my own.
But sometimes, that's bad: We don't live in a perfect world. Enemies exist. Things like ITAR can't prevent a sufficiently smart person from doing anything, but they do make it a lot harder for them to get started.
correct. and the idea here is to put up at least some form of barrier to entry, because any good guidance is also nuclear warhead delivery guidance at a certain point.
I launched a high altitude balloon as part of a summer school program over a decade ago and we checked on edge cases. Off the shelf GPS are supposed to not work beyond a certain height 18,000 m and/or speed 515 m/s to be a barrier for use as a weapon. Some hardware treat that as AND; some treat that condition as OR. The term to look up is “CoCom Limits”.
Nice reverse engineering! As a hacky way for the non-tech-savvy, couldn't you use a temp account to create ticketmaster account and then buy the ticket and then sell the temp account information to bypass their rules?
This reverse-engineering also breaks if ticketmaster forces venue staff to only scan if the barcode is in the ticketmaster app. Unless you create a lookalike app to trick the staffers.
I am not an expert, but I think one of their layers of protections (that is, to ensure that TM itself gets the greatest share of scalping money) is applying much greater scrutiny to freshly-created accounts when it comes to the in-demand events. I'm not sure how they effectively bootstrap new legit users of course, but I've been offered I think around $100 to sell my Ticketmaster account, which is old. (I can't recall how they found me, perhaps it was an ad just stating that they'd buy an account older than X years).
Phone number? The friction/expense of a scalper getting a new one for every sale would seem sufficient. Although I guess the scalper could reclaim (via password reset or whatever) accounts after the show to some extent.
Good luck forcing a check like this at a busy event venue.
I once paid at Starbucks with the Apple Wallet barcode appearing in a photo of my phone displayed on the back of a DSLR. Plopped my not-remotely-iPhone-like Nikon D800 on the counter lens-down, LCD-up, barista scanned it without a second thought.
The article and comments debating best practices in notebooks remind me of the time I hosted a website from a jupyter notebook.
Technically, I was using gradio to create a localhost webpage and then piping it through cloudflare. The website would only work when the notebook was running on the cmd line of EC2. Hey if it works, it works! Notebooks allowed me to do a 2 week project in 2 hours.
Like most things, notebooks have their pros and cons. One of the biggest adv is very rapid experimentation. Even a regular script takes a while for python interpreter to run and that time (even if only a few seconds) adds up in lack of creativity ("Bret Victor - Inventing on Principle" [1]). And if your script is loading a big database, then notebook is a no-brainer.
One of the biggest disadv of notebook is mis-ordering. You are allowed to declare a variable in cell 3 and then go and use it in cell 2. Even worse, you can declare a variable in cell 3 and then delete cell 3 and still be able to use that variable. That I believe is the biggest dis-adv of notebooks. It adds way too many subtle errors. One way to bypass this is to write everything in functions - no global vars.
I am willing to accept the issue of mis-ordering in order to get rapid experimentation. It's subjective whether you think the pros outweight the cons. I definitely think they do.
"Interestingly, the only thing stopping civilians from using the P code is the knowledge of the value of its chipping sequence. If the formula to generate the P code was publicly known, there’d be nothing stopping civilian GPS receivers from locking on to it, with the exact same techniques as are used for the C/A code."
I didn't finish reading the whole thing but was curious. Is there any way of brute forcing it or some other trick to get the chipping sequence to get the P code for more precise GPS?
Kind of. The P code itself is publicly available in publicly available specification IS-GPS-200. They are xor-ed by W code, which like the sibling comment says, is cryptographically generated, which runs at a lower rate. This results in P(Y) code signal, which is then transmitted over the satellites.
If you know whats the P code roughly is, which you can know from L1CA code and handover words, you can try using both possible W code value (its 0 or 1) and that W code practically dissappear, until the W code move to its next iteration and you need to guess again. This strategy is called semi-codeless tracking.
People did this so they can access the signals at L2 frequency (1227.6 MHz), which because its at a different frequency than L1 (1575.42 MHz), they got slowed down by different amount through the ionosphere. So by comparing the signals together, you can remove the ionospheric effects and get more accurate position.
These days you don't need all of this, GPS now have public signals at L2 and now smartphones have these fancy dual-frequency receivers
I believe you are correct. My understanding is that the root P key is rotated daily, and needs to be manually uploaded to any military hardware that needs to use it.
"Interestingly, the only thing stopping civilians from using the P code is the knowledge of the value of its chipping sequence. If the formula to generate the P code was publicly known, there’d be nothing stopping civilian GPS receivers from locking on to it, with the exact same techniques as are used for the C/A code."
I didn't finish reading the whole thing but was curious. Is there any way of brute forcing it or some other trick to get the chipping sequence to get the P code for more precise GPS?
Does anyone have a comparison of Airtag vs "Bluetooth tracker tags from Chipolo and Pebblebee"? I imagine Airtag is superior due to non-reliance on Bluetooth? Can Chipolo/Pebblebee tap into the Android network without Bluetooth? And is their tech comparable to Apple's work?
In addition to BLE, AirTags use UWB. I assume this is how AirTags can provide fairly high-fidelity directional guidance, something that wasn't clear the new set of Android tags provide.
I can't really see how Airtag can be superior in most places around the world: it's simple, there are many more Android users than iPhone users so the Google network can only be better (if not right now, then over time, give it 12 months...?)
This difference is even stronger when you leave EU + US and go to Asia, Central / South America, Africa where iPhone users are not less, but very rare.
My use case is clearly to track my hand luggage, wallet, backpack etc. when I travel and I'm a nomad. Gonna wait on reviews but I'm very excited!
What if you cover the drone with non-reflective material? Or does it care about the reflective nature of the core of the object rather than just the surface?
Frequency is relative to the size of the object you want to detect and the size of the wave. Higher frequencies detect smaller things and can discern between them. Go too far and you detect bugs. Too low and you only see barn doors.
Spinning blades are a dead giveaway for radars. Certain things reduce radar signature but the interesting part now is electronic countermeasures.
A radar can produce a fuzzball graphic to the viewer of the data, speed, distance, direction of movement, etc.
I read the article but couldn't find any mention of how a radar detects drones? A drone is much smaller than an aircraft, can be confused with birds and can have 0 communication with a command & control station due to AI. So how can anything detect it?
Plus, even if you detect it, it's coming towards you at a speed of 100+ kmph with the intent to crash into you and detonate the payload. Any missile you use is way too expensive relative to the price of the drone. So what to do?
I say this bec the drone footage coming out of Ukraine is shocking. I saw a video of a drone just following a soldier for 30 seconds while the soldier was trying to run away from it. The drone crashed into the soldier and exploded. That is absolutely black-mirror dystopian stuff.
I wouldn't be surprised if the drones they can detect are the larger fixed wings long range ones and not the FPV types.
FPV drones, aside from EW and directed energy weapons, don't have any counter at the tactical scale (although fielding shotguns might help). Any counter will need to be infantry section/platoon scale to be effective, or at a minimum company level weapon system. The range of the AA countermeasure for an FPV drone will be on the order of 500m-1.5km, so covering a 1000s of km long front line is near impossible with current systems.
Even if the entire front is utterly uneconomical to cover, local defenses like these can be used to provide an umbrella to more important points. Field hospitals, ammo dumps, headquarters, maintenance depots, etc.
I understand I am making assumptions bec of lack of information in the article.
Re shooting them down, if you can't shoot them down, whats the point of detecting them? And a drone being kamikaze only matters IF you can shoot them down. Because then it's a race between the drone rushing to kamikaze vs you shooting it down.
I will admit that I am just speculating here because I have a strange interest in militarized drones and little technical knowledge.
PS : Defending your position by saying that a company doesn't fuck around is not a very good argument.
You don't handle a dji dropping mortar rounds the same way you handle a fpv drone with a rpg warhead strapped under it nor the same way you handle a fleet of bayraktar pounding you from 5km above. I'm just saying you're mix and matching things which shouldn't be mixed and matched, the premise is wrong, no one is going to shoot down a $50 dji with a missile and no one will try to shoot a bayraktar with a shotgun. Just like you don't shoot a tank with a rifle or a moving plane with an unguided rocket launcher, you need the right tool for the right job
You can scramble them, shoot them with bullets, catch the drone/projectile with nets, shoot them with missiles, &c.
Again it's one of the main military company in France and it is partially owned by the state, I can assure you the fact that these things are being made is proof that they fulfil their role
> Any missile you use is way too expensive relative to the price of the drone.
My understanding is that you use a laser, though I don't see why even regular AA guns wouldn't work, especially if it is trying to crash on you, meaning it needs to get very close.
> Plus, even if you detect it, it's coming towards you at a speed of 100+ kmph with the intent to crash into you and detonate the payload.
That will at least help you distinguish it from a bird :) Plus, I suspect that in any military conflict, you'll be more than happy to eat the cost of frying a few birds if it means you stopped the enemy's drones.
> Any missile you use is way too expensive relative to the price of the drone
Western missiles have been designed for generations assuming they're aimed at something expensive you want to break, or to break something next to something sensitive you'd rather not. We have the technology to create swarms of well-enough aimed small rockets. There simply wasn't a niche until now.
This has a range of 400+km (100kmph = 4 hours to respond).
It's purpose to maintain an accurate picture of the sky, tracking many objects at the same time at different altitudes. The radar itself is just part of the picture - its data is integrated with command and control, where decisions are made about the picture coming in, and the data is blended with other sources to get a more accurate picture (not sure if bird or drone? check the video feed data, or a number of other sources).
The unit itself looks at the behavior of the object (speed, acceleration, routes, elevation, radar profile) to determine the likely class of the object (birds don't fly like drones nor do they have the same radar profile).
As far as mitigating threats, command and control again makes those decisions. It depends on the context of the fight and the resources available. In the hypothetical situation you think a kamikaze drone is headed your direction, the radars are mobile - one option is to simply move. You may know (or have a good guess) as to the specific threat - how it is controlled. You might take out drone communications with EW. You might misguide a precision munition by spoofing GLOSNAS/GPS so that it drifts and misses its target. If it is flying at a low elevation, you might be able to take it out with heavy machine gun fire. You might decide to let it strike, due to cost-benefit ratio.
It really gets down to specifics: What's the air asset? What the threat? What's the mission? What's the battle context? What are the resources?
Regarding a drone having zero command & control due to AI - most of the "AI" in drones is simple straight line flying for a couple hundred meters (so called "terminal flight guidance"). This is because enemy electronic warfare cover may jam communications channels for the drone as it approaches closer to a target. As cool as it sounds to have fully autonomous drones making complex decisions, piloting around obstacles in all weather conditions in 3D space, tracking moving targets, etc - this isn't the threat from drones right now.
1) Small drones can be detected. I'm not gonna go into the specifics, but Orlan sized drones with 3m wingspan can be detected from hundreds of km. Smaller sized commercial drones, which are 1/10th of that size, can also be detected pretty well.
2) No mater how small these drones are, they are dependent on some nav and coms system. Even autonomous "fire-and-forget" drones need a somewhat robust GPS link for navigation. For operated drones, any telemetry can and will be linked to.
But, ok, let's assume some futuristic drone that has a powerful AI system to do all its navigation via onboard sensors, which do not transmit or receive any information. How could such a drone get past a radar system? By either being too small for the radar to detect, fly too low for the radar to detect, or have some geometry that voids detection. Or the radar gets jammed, while the drone tries to get past it.
3) Drones have features which can be detected by radar. Motors, for example, would be one of those.
4) Radars are rarely the only sensors used. You have a whole array of different sensors which can be used to pickup stuff. Even with the radars themselves, you could have one radar for detection / target acquisition, and another radar for precise imaging.
There's no free lunch, though. A very small drone would mean limited range and payload, which in turn means you'll either have to deploy it close to the enemy, or via some larger craft.
Flying a drone too close to ground ads tons of interference to the drone, not to mention detection by things like acoustic sensors, humans, cameras, and what not.
But that also goes for the radars. Small targets can easily disappear in clutter, or dip under the elevation of the radar.
Admittedly I do not know much about drones, but the point was just that any telemetry that goes out of the drone, is relatively easy to pick up - and is how the military track drones.
At least today the vast, vast majority of combat drones are operated - and the video feed is one of the first things that get picked up from them.
These videos are horrible. On one side I feel for that kid who probably didn't chose to be there and grappling with imminent death. On the other hand if my country was invaded I would probably watch those videos with popcorn. It's a weird war, mixing social media with a ww1 trench war, military hardware with a share button, you get to watch live the horror of the trenches from under your duvet.
The other thing is those drones are mostly made of plastic, have no hot exhaust. Every time Ukraine is glad they shot a $50k drone with a $1m patriot missile I wonder where this war can possibly go.
Patriots aren't the answer to Shaheds, but that's the wrong way to look at it. You don't compare the price of an APS charge with the price of an RPG. You don't spec your soldiers' ballistic plates by aiming to bring their cost down below the cost of the bullet you hope it'll stop. The plate is replaceable, the soldier's heart isn't, you sacrifice the thing you're happy to lose to save the thing you want to keep.
They're not happy they shot down $50k of drone, they're happy they shot down 50 kg of explosives that was going to strike a power station, it's not a hard trade to make.
A key difference from your examples is that the enemy doesn't have the capability to simply (and risk-free) send an arbitrary number of bullets at your soldiers' hearts or RPG rounds at your vehicles; the major "cost of delivery" is not the bullet/RPG but the trooper (and his risk) bringing the gun to get a chance to make a targeted shot with the bullet/RPG.
If bullets could leave the factory and magically fly straight at your soldiers, expensive single-use ballistic plates would not be a practical solution - you'd simply run out of them as you can't possibly produce as many of them as the enemy can make bullets.
>On the other hand if my country was invaded I would probably watch those videos with popcorn
We don't even know which country the soldier belonged to. And realistically, if your country was invaded, you are unlikely to be in a position to relax and eat popcorn but I get your sentiment.
I agree how weird this war is. The first war being live streamed. That increases the horror since watching videos of something happening is much worse than just reading about it. And most people don't read but will intently watch a drone following a soldier and exploding.
> Every time Ukraine is glad they shot a $50k drone with a $1m patriot missile I wonder where this war can possibly go.
More drones and more missiles. Until someone figures out cheaper ways of stopping drones. My gut feeling is that in 1-2 years, we will have proper industrial ways of stopping drones en masse. And then drones will become just a part of the commander's arsenal like any other weapon.
> And realistically, if your country was invaded, you are unlikely to be in a position to relax and eat popcorn but I get your sentiment.
Most military age ukrainians do. Ukrainian mobilisation is nothing of the scale of ww1. France was roughly the same size as Ukraine in ww1 and had 7m men under arms. Ukraine has less than 10 times less.
All the decent open-source ones are fairly basic with limited fine tuning and no alignment (RLHF).
I plan on adding those things. Although I am not sure if there will be any demand for it. Plus, there's a decent chance meta will make llama 4 speech output making this one obsolete.
reply