Hacker Newsnew | past | comments | ask | show | jobs | submit | 14u2c's commentslogin

> DeepSeek can surpass proprietary models by “profiting” from open research and open source, why couldn’t the proprietary models do the same?

DeepSeek did something legitimately innovative with their addition of Group Relative Policy Optimization. Other firms are certainly free to innovate as well.


Doesn't seems so surprising to me. Most school shootings are perpetrated by those involved in the school community in some way. Similar domestic violence, humans are a lot more likely to commit these types of crimes against those they know personally.


In Southeast Asia? The US military probably didn't even hear about it until after it was missing for a bit.


As is often the case almost anywhere in the world, there's a major US military base nearby. In this case it's Diego Garcia, and is absolutely huge:

https://veteranlife.com/military-history/diego-garcia/

"The base is home to thousands of American troops, sophisticated radar, space tracking, and a communications facility."


> it's Diego Garcia, and is absolutely huge

You know what’s huger? The Indian Ocean. Add to that the lack of blue-water competition in it and it’s totally reasonable that any birds we have in the area are craning their necks north, not down.


“Nearby” is doing a lot of work there. The presumed flight path of MH370 went tens of thousands of kilometres from Diego Garcia, certainly outside the range of even the latest over-the-horizon radars. Meanwhile the tracking stations on that base are mostly focused on space activity and missile launches.

No amount of money can break the laws of physics and, like any sensor, radar has fundamental limits in range and resolution. MH370 flew into one of the most remote parts of the ocean it’s possible to reach, it would’ve been a miracle if it was tracked by any radar.


The presumed flight path at the time it went missing is perhaps relevant, and that was within the vicinity of Diego Garcia, as the flights last few known turns were heading toward it. It seems odd the US would not care nor have tracking ability.

I've seen analysis that showed the flight was within range of several over-the-horizon radars at the time of disappearance and for hours after, ie - someone should know more than we do. As often with such things, I can't find it again.


OTH radars are not operating all of the time, given the cost to operate them, and are not just covering large areas of ocean all over. They're typically focused specifically on areas of most importance which, for Diego Garcia, would be north toward China, not East toward Malaysia.

Nor do OTH radars always operate at maximum efficiency: They achieve their longest ranges by bouncing signals off the ionosphere, which is severely affected by prevailing space weather.

The only radar that it likely did pass through was Australia's JORN, but the western sector was not operational that night and isn't on 24/7 because of cost constraints.


Good points but neither of them rule out being reconfigured and used in an emergency, potential hijack situation to locate what could be a significant security threat. Airliners were used for the biggest attack on US ground since WW2. Priority number one.


Only if there’s enough forewarning, the radar is operational, the aircraft is within its range given prevailing space weather conditions, and that it’s pointing in the right direction. The latter is important because OTH radars are almost all fixed and can’t be steered.

Yes, governments would love to have global 24/7 coverage even over the open ocean. In practice that’s neither possible nor practical.


> OTH radars are almost all fixed

Given their strategic advantage, you can bet the military will have prioritised steerable.

I disagree with the general premise regarding global coverage. With the US military, capabilities, especially in surveillance, have historically been shown to be decades ahead of what the public thinks is possible.

Personally, if a post appeared tomorrow showing some HN had figured out how to trace the movements of any airliner, without using its adsb and relying instead on anything and everything else that's publicly available, from satellite imagery, to radio frequency data, to radar, even weather data (contrails are often detectable), it would seem cool, sure - but, not unbelievable.

That hypothetically believable scenario would be one person, with no budget, in likely a few weeks or months of their spare time.

The US military has trillions of dollars, the best talent in the world, and decades of dedicated effort in exactly this area, and a propensity to keep such advances secret for decades (as shown recently enough by the Trump photo).


> Given their strategic advantage, you can bet the military will have prioritised steerable.

This is not something you can really prioritise. OTH radar designs are a trade-off between range, angular resolution, frequency, and mobility. For the longest-ranged systems with good angular resolution you can't steer them outside their set beam pattern, because their sheer size makes that kind of steering impossible. So if you want steerable radars you necessarily have to compromise on range, angular resolution, etc.

> I disagree with the general premise regarding global coverage. With the US military, capabilities, especially in surveillance, have historically been shown to be decades ahead of what the public thinks is possible.

Again, there are fundamental limits here. As much money as the US military has, it can't break the laws of physics. We also have a good sense of what types of assets it has and where they are, including satellites.

> Personally, if a post appeared tomorrow showing some HN had figured out how to trace the movements of any airliner, without using its adsb and relying instead on anything and everything else that's publicly available, from satellite imagery, to radio frequency data, to radar, even weather data (contrails are often detectable), it would seem cool, sure - but, not unbelievable.

Doing so over the vast open ocean would indeed be unbelievable. Even doing so for an individual over an ideal location would not be believable, as available resources don't make this possible at any real scale with the necessary granularity.

> The US military has trillions of dollars, the best talent in the world, and decades of dedicated effort in exactly this area, and a propensity to keep such advances secret for decades (as shown recently enough by the Trump photo).

See my point above. As for the Trump photo, by which I presume you're referring to the satellite image of the failed Iranian launch, the displayed resolution was within what experts had already presumed was within the capabilities of deployed US satellites given all available information. The photo didn't display surprising capabilities, it merely provided an official confirmation about what was already widely assumed.


What's expensive about operating radars?


At this scale:

1) Electricity usage

2) Staffing costs, both for operational control and ongoing maintenance and support

3) Parts replacement costs. The more you use the radar the faster its components wear out


Thanks for the reply.

I'm doing research on radars at a lab, but on automotive and robotics scales and operating costs are just not something we consider.


No problem at all.

I can imagine so. For context, OTH sites are generally massive with hundreds of large antenna elements. This article has some pictures of one of the JORN sites as an example: https://www.australiandefence.com.au/defence/cyber-space/jor...


Not tens of thousand of kilometers. Diego Garcia to tip of Indonesia is 2800km and 3000km to Australia. But the rest of your post holds.

Also, the over the horizon radars are in Australia. I think they could have picked up the flight but were offline that day.


Appreciate the correction, I meant thousands.

Yes, that’s the JORN system. The western sector might’ve picked it up if it went near western Australia as presumed, but it wasn’t operating.


Buy a $50 inverter. They are tiny and it's not like this thing is using enough power for the efficiency hit to make a difference.


It's absolutely significant in at least some of the situations where you'd use this. A friend and I are planning on taking a small sailboat across the atlantic next year and were looking at communication options.

We had decided the normal starlink is just not feasible on a small solar-only boat so this caught my attention. It's still too high though. It would take slightly more power than all the other systems combined if it were DC. With an inverter it pushes it over and there just isn't that much prime panel space on a sailboat.


Iirc the hit from ac/dc conversion is 10-20%? So excluding ac from the equation saves 20-40% power use


Except that this was a fixed price contract. If they don't get it right the pork train does come to an end.


There will always be more fixed price contracts in the future.

Every time the barrel emptied in the past, a new barrel full to the brim with pork was delivered.


That bsd analysis. There isnt a magical contract fairy that creates infinite contracts for every company.

Yes there are more fixed price contracts, but if you fuck up like Starliner you are less likely to get another one.

And even if you could get it. Boeing peadership has clearly stated they are basically not really interested in that anymore. The have lost billions on fixed price contracts in space and military.

Boeing has not been able to get much new from NASA. Their moon lander was basically embaracing. They are in the nee private station buissness much.

One SLS gets finally mercy killed they will not be a prime contractor anymore.


> magical contract fairy

I've never heard anyone call the federal government that.


In a pre-SpaceX world, there pretty much was no one else they could lose the next contract to. I think a lot of Boeing still lives in a world where they have a good reputation to milk.


This would make sense if the money was paid upfront.

It wasn't. As of right now, Boeing is massively in the hole on the project. The only way they get paid if they successfully complete the development program and then fly the 6 operational flights.


That doesn't seem to actually be the case?


It is the case. Boeing is losing billions.


Most of the big "investments" in OpenAI are in the form of compute credits. I fail to see the downside of that.


Nope, it's just a regular ATM operated by a 3rd party company. You get cash from it then give them the cash. The store will also often reimburse for the ATM fee.


Nope, see other comments. My interpretation is correct.


Because they still pay $500k cash.


The thing with Netflix is that they have such few open roles and there are plenty of other liquid high comp companies that pay similar. It was all about the original stock growth that just stuck I guess


This highly depends on your skillset. Most employers with equivalent liquid comp will be HFT, and you will be writing C++.

Netflix is the only employer I know of where an Android or iOS IC can make liquid 500k+ TC without RSUs.


What's IC


individual contributor


What does "cash" mean in this context, as opposed to if the word wasn't included?


They famously don't pay you in equity at all. If you work at Netflix and want NFLX stock, you buy it yourself.


> (1) there are only about 3 big browser vendors in the world, and even Firefox and Safari are lagging in features and bug fixes

Are you forgetting WeChat? Because that's all they likely care about.


The X5 browser core in WeChat seems to be based on Chromium (and was vulnerable to Chromium-only vulnerability CVE-2023-4357)


Do they ship their own browser?


WeChat is a 'super app' to the point that you don't need really a browser in China anymore. It's a mobile first based ecosystem (WeChat, XHS, Douban, Taobao, etc)


I know that. But is there no web access at all?


Why not use something like Cloudflare for DNS? You'll never have to deal with the registrar again after setting name servers.


CF has such a limited TLD selection. I know it's growing all the time, but sites like namecheap seemingly support every TLD as soon as it comes out.


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: