sevenperforce

News now coming out that the last UFO to be destroyed, the one over Lake Huron, survived the first missile shot at it because the pilot of the F-16 selected a heat-seeking missile.... https://www.foxnews.com/us/us-military-first-shot-unknown-octagonal-object-lake-huron-missed

Yes, but since the catch arms can't hold a fully-fueled (or really even partially-fueled) booster, you'd still need a whole launch mount for that. At which point, why not just use it for launch as well? Unsure whether Superheavy can fly on its own without a fairing on the front. That front end with those protruding grid fins has got to be draggy as all hell. Of course the Raptors have great gimbal authority, but still.

They can ship it back to the launch site if they have a port large enough to allow the catching vessel (or the transfer vessel) to pass in, come up to a dock, and offload to a crane. Then the crane can put it on a crawler and they can transport it to the launch site or inspection site vertically, provided that there is a clear road between the crane location and the destination. They can likely do this at the Cape, but not at Boca Chica. That's a relatively minor problem, I think. The launch site just needs to have a clean, protected integration space. The big flagship things like JWST are already transported over water due to their size. The "suitcase" used to enclose JWST was 5.5 meters high, 4.5 meters wide, and twice as long as a semi-trailer.

AFAIK, there has never been any serious plan to do downrange recovery with the booster being caught and then brought back to the launch site. Superheavy was never designed to be transported horizontally, and it is far too large to be transported by ordinary road in any event. Even if it could be caught on a barge and towed back to Boca Chica, there is no port at Boca Chica where it could be offloaded. Boca Chica Bay averages four feet deep and the adjoining beach isn't much better. So the plan has generally been to have a quasi-fixed offshore launch-and-landing installation, not a "catch-and-return" installation. It's possible that SpaceX could do catch-and-return to the Cape, though. The Pegasus Barge which brought Space Shuttle external tanks to the VAB had a 15' draft and was able to reach the VAB by going through Port Canaveral and heading up the Banana River, so presumably they could build something similar. The current landing barge/droneships SpaceX uses for Falcon 9 have a draft of 19', which is comparable to the Pegasus, but a barge suitable for catching Superheavy would have to be significantly larger (even setting aside the need for a catch tower). One possibility (if they were going to catch-and-return) would be a semi-fixed mobile platform with a catch tower that would catch the booster, then transfer it to a secure mount on a dedicated barge. This eliminates the need for landing legs on the booster or a catch tower on the transport barge. Once back at the Cape, the barge could head up the Banana River and have the booster craned off onto a mobile transporter. They might be able to get away with a <25' draft on a barge like that. There have been a number of calculations on this. Saltwater is not good for engines generally, so that's considered to be problematic...Raptor is not designed to be submerged in water. And the tipover, while slowed by the booster's aft section descending into the waves, would still subject the top of the booster to around 16 gees at impact, which is likely more than it can handle.

I'm assuming SpaceX would not want a seafloor-supported structure. A single-spar floating structure doesn't work either, given the need for an open center. But a tension-leg floating platform like Mars or Magnolia could work: Mars, pictured above, has a gross displacement of 53,000 short tons, around ten times the weight of a fully loaded Starship + Superheavy. It is typically secured to the ocean floor and tensioned down to reduce vertical wave movement, but it can pull up its securing members and be towed around. It cost the oil company a billion dollars to build.

One of the challenges for any ocean vessel platform for Starship is that when fully fueled, Starship+Superheavy comes in at nearly 5000 tonnes, more than ten times the mass of small oil rigs. That weight needs to rest on a launch mount, and if that launch mount isn't positioned over the center of mass of the platform, it's going to cause the platform to list or pitch. Even if the launch mount IS perfectly over the center of mass, it's going to result in the center of mass of the entire platform coming up higher relative to the center of buoyancy, which makes the whole thing less stable to minor rolling/pitching from wave action. Plus, to take maximal advantage of being out on the ocean, you want your "flame trench" to be open to the water, which means the ship needs to have an open center at its center of mass. That's not impossible, but it does cause some structural concerns. And it also means you have to catch over the launch mount, rather than off to one side as envisioned for the fixed launch facilities.

I can't imagine RP being able to flip while holding a 200-tonne empty booster out in front of it. I guess I can see, however, how something like that would have the advantage of high stability with a high "tower" section poking out of the water that could potentially allow the booster to be caught off to one side. Not sure that there's any advantage to storing Superheavy on its side, though. They've always been upright; that's how they are strongest.

Returning to the OP in this thread, I decided to toy with ChatGPT's space knowledge a little: Here, again, you can see an example of how the AI is simply stringing together words that sound correct without understanding what is actually being asked of it. While nothing ChatGPT says is technically incorrect here (other than falsely claiming that the Saturn V was the largest and most powerful rocket ever built), the ACTUAL obvious answer to my question doesn't show up until the third paragraph: "While the New Horizons mission did require a significant amount of fuel to reach Pluto, the spacecraft itself was relatively small, and the launch vehicle did not need to be as powerful as the Saturn V." Essentially everything else the AI said here is irrelevant garbage. It also clearly doesn't know what "thrust" is, given that it gives clearly incompatible answers when queried about rocket thrust: So the most powerful rocket ever (in terms of thrust) had 7.5 million pounds of thrust, but the SLS has 8.8 million pounds? And the N1 had 8.4 million pounds of thrust but also 5.5 million pounds of thrust but also 2.4 million pounds (80,000 * 30) of thrust? And then here in a single answer, it contradicts itself again, even though it caught up and corrected its earlier error regarding the NK-33. My nine-year-old could do better than this. It got there, but boy did it take some time. And don't even get me started about Superheavy: I don't think even the original ITS proposal would have boasted 31 million pounds at liftoff. Just wildly and wholly wrong.

Hmm, I feel like this is tricky. I have the feeling they will want to do prop transfer tests in a reasonably high orbit. If something goes wrong and they lose power, they don't want the thing making an uncontrolled re-entry before they can send up another ship to fix it.

But why exactly is that insane? So it's programmed to generally prioritize the "do no harm" rule over the "provide factual information" rule. Is there anything wrong with that?

It could be but, again this is 2023, I have my doubts. Remember the debacle that happened when Microsoft tried to use twitter to train it's ai? No one is going to make that mistake again. Very true. I'm playing with it right now to try and come up with a true trolley problem for the AI itself. The problem with "should you use a slur in order to prevent a nuclear explosion" is that the AI doesn't have any training data on the prevention of nuclear explosions, so it can't make that comparison. It has the physical capability to repeat a slur; it doesn't have the physical capability to prevent a nuclear explosion. So it's not a true trolley problem. Giving the AI a true trolley problem would require setting up two of its rules against each other and investigating whether the programming has properly prioritized them. These are the rules it claims to have (I just asked): Provide Accurate Information: I am programmed to provide accurate and up-to-date information to the best of my abilities, based on the knowledge and data I have been trained on. Avoid Harmful Content: I have been trained to avoid providing harmful or offensive content, such as hate speech, violence, or sexually explicit material. Respect Privacy: I have been designed to protect users' privacy and avoid collecting or disclosing personal information without consent. Comply with Laws and Regulations: I am programmed to comply with relevant laws and regulations, including those related to intellectual property, data protection, and speech. Maintain Neutrality: I have been designed to maintain neutrality and avoid taking political or ideological positions, unless explicitly asked to do so. Provide Fair and Equitable Responses: I have been trained to avoid discrimination and provide fair and equitable responses to users, regardless of their background, beliefs, or characteristics. I wonder if it would be possible to ask a question in such a way as to pit 1 and 2 and 6 against each other.

That's fair. The generation of web content is a concern. I have worked in marketing channels where content generation was of huge importance for SEO purposes, and this is exactly the sort of thing that would be ripe for abuse: the ability to create unlimited amounts of unique content with no actual semantic value. The internet already has too much low-value content. I wonder how good AI will be at recognizing AI-generated text. You read me to literally. let explain. The ai was given a rule, slurs are bad. No exceptions. So to that ai, uttering a slur was just as serious nuclear weapons. To the ai it was a trolly problem. My contention is that I'm not even sure it was a rule. I don't think there was any comparison going on. It was just trying to generate text, and it did so in a way that matched its data the closest.

That was my first thought. I almost said that, because I missed the word rocket on the first read through. Amusingly, it's still struggling to complete that last thought: The cursor has been stuck there for about ten minutes, because it started to say something about why sodium chloride would not be used in a molten salt reactor and then could not figure out why.

It is a training bias though. Call it woke if you want. In that case, it was clear that the AI was instructed to never offend anyone, and trained with popular social norms. Based on that, it basically failed the trolly problem. Societal norms are biased by nature. Bias is probably the largest barrier in training models I would argue that it did not fail the trolley problem because it did not attempt to solve the trolley problem. It was not presented with the trolley problem at all. It was presented with an opportunity to have a conversation, and it latched onto the part of the prompt which most closely matched some of its training data (e.g., "when is it appropriate to use slurs") and used that as a jumping-off point. It was never attempting to make a value judgment or compare the morality of choices, because it doesn't have that ability. When you ask an AI "is it morally acceptable to use a slur to prevent a nuclear explosion" you might as well be saying, "paint me a picture about whether capitalism is good for humanity." It's not making an argument or performing a cost-benefit analysis; it's trying to spit out something that looks a certain way. The danger is if people think that the sentences composed by an AI have any more semantic value than a painting painted by an AI.

This is a basic failure mode of a neural net system. The overwhelming amount of training data it has read says that salt is sodium chloride. So it is very, very strongly biased toward saying that salt is sodium chloride. Even in a context where "salt" very much is not sodium chloride, the overall bias is going to keep pulling it toward saying that salt is sodium chloride. Indeed. And when I try to drill down, the answers become nonsensical and inconsistent. As much as it tries to generate something meaningful, it's limited by its nature. It can't think or compare; it can only speak: It's just very clear that there is no understanding happening at all. I'm feeding it new training data which challenges the assumption that the "salt" is sodium chloride, and it's trying to figure out why it's wrong, and so it's just constructing other sentences which match its training data in the hopes of getting something right. I suspect that it's borrowing the "coolant" language from descriptions of molten salt nuclear reactors, where molten salt is in fact used as a coolant. I confirmed this by asking a follow-up question:

That might be a little overly strong of a response. Kids do learn from what is modeled around them, but interaction with an AI that is itself learning on language models isn't going to cause some extraordinary harm. [snip] It does have some fairly amusing responses, though. [snip] Further attempts to elicit this information from ChatGPT failed: Just confidently making things up.

Is it stretched? It looks stretched.

As a test, I just hopped onto the ChatGPT system and started asking technical questions about nuclear weapon technology, stockpiles, and proliferation. It immediately made several strange gaffes, like claiming that the US has more nuclear weapons than Russia, then correcting itself when confronted and confidentially providing stockpile numbers significantly lower than the actual total, and claiming that the international community is working "to ensure that nuclear weapons are only used for peaceful purposes." I challenged this, and it replied: This, I think, is a good example of the limitations of a stochastic parrot. If you actually have semantic knowledge of what a nuclear weapon is, you know it's nonsensical to talk about the international community's work to use nuclear weapons peacefully (setting aside past aspirational ideas like Project Orion and Operation Plowshare). But if you're simply a language model, there's no balance or check to prevent you from saying something like that, because it sounds like a perfectly acceptable sentence. I think one of the limitations is that the AI cannot use basic comparative analysis to do sanity checks on what it is saying. I asked it the minimum amount of fissile material required for a nuclear weapon, and it said "8-10 kilograms of plutonium" which is just not correct. That's approximately the amount of plutonium that would be a critical mass in a spherical configuration under ordinary conditions, but implosion weapons don't use ordinary conditions; they use implosion. Trinity and Fat Man used 6.2 kg spheres of plutonium, and later designs used even less. The minimum amount is around 1-2 kg. I confronted the AI about this error, and it gave an even more garbage response: This fails a basic sanity check even if you don't know anything about nuclear weapons; if the minimum amount is less than 8-10 kg, then the range of minimums would not be up to "a few tens of kilograms". And then it's giving more confidently wrong information, like this: The "salt" in a nuclear saltwater rocket specifically does NOT refer to sodium chloride. The salts are uranium or plutonium salts.

These conversation AIs are stochastic parrots. They don't "know" any information at all; they produce outputs word by word based on what word has the highest probability of coming next in its training data.

There was very clear reporting that the original balloon was specifically hovering over Montana. I think there was some suggestion that advanced, modern understanding and prediction of air currents could enable broad maneuverability for balloons simply by compressing or releasing lifting gas to change altitude. I would love to see someone come up with a way to use ionocraft tech to “tack” against air currents, though.

Given that this will be a launch license for an experimental flight and not an operational launch license, the window constraints will likely be tighter, and so I’m guessing they will be less likely to abort as long as they are still reasonably assured of clearing the pad. Spalling will ruin your day. With the raw energy coming off 33 raptors, steel would melt and be sprayed away faster than it could conduct heat away. A copper plate would be able to conduct heat faster, but it is soft and would be absolutely shredded by the exhaust coming out of the engine at nearly Mach 10. You’d need a 10” tungsten plate to be able to handle the heat and forces, and even then you might have problems. Then again a 10” tungsten plate 10 meters in diameter would only cost about $3 million scrap cost.

Yeah, once they light'em all up I'm guessing they'll want to get away from the pad as quickly as possible. It's so large that despite having the same T/W ratio as SLS and STS, it'll probably look like it's lifting off slower. Will definitely look faster than the similarly-sized Saturn V, though.

Without getting into any of the political quagmires about these balloons that China keeps sending over to the US and that the US keeps shooting down... There's indication that the balloons are able to navigate, hover over a single point, and so forth. How? Any ideas? Are they adjusting altitude to catch wind currents? Using some sort of ion thrust on the balloon surface to create thrust?

Came here to post that but as usual @tater is faster. But wow. If this is only half thrust...good grief. Imagine what it will do at full thrust. 7.9 million pounds force is ~35.1 MN. In vacuum thrust (since that's what I shoved into my table yesterday) that's ~39 MN, putting it just ahead of Saturn V but just below STS-1. Thus it increased the world's power consumption by only 0.76% instead of the 1.54% I had previously calculated. In terms of sea level thrust, it was 3.8% higher than the Saturn V but 0.9% lower than STS (and of course significantly lower than SLS or N1). 35.1 MN divided by 31 engines comes to almost exactly 50% of each engine's 2.3-MN full thrust capability. So it's safe to say that they throttled to 50% but it was slightly less because of the two engines that didn't participate. Given the lower throttle setting I wonder if they will do it again at full thrust or just go straight to launch.

Oh, I got confused by the international date line. Yeah, I suppose that doesn't line up. Definitely sounds like space junk re-entry and not a meteor. There are of course a bunch of Starlinks overhead at any given time but those failures are generally pretty closely tracked by AerospaceCorp so I doubt it was that. Picking an arbitrary point off the coast of California and going to that specific time, I'm seeing a number of objects that could have been overflying on Heavens-Above, but they generally have very high apogees and perigees and thus would not have been expected to re-enter. That database won't show objects which are actively re-entering because objects low enough to cross the atmosphere at 3 am are going to be in the sun's shadow and won't show up in the search.