jjwiseman

First, OpenAI is not claiming that the model wasn't trained on those sessions. What they've said is “We (the researchers and the agents) did not see any of their work through any means until they released it publicly — in particular, no specific user data was accessed in order to solve this problem.” and “We did not use their prompts or proofs to prompt our models or direct our agents.” and “While unlikely, we cannot rule out that de-identified data derived from their usage of our products helped improve our models.”

They also said “Our effort began on September 1st after hearing a rumor which we later realized was related to Levent Alpöge … and Tristan Buckmaster….” They say the rumor was that two Millennium Prize problems had been resolved, and that this prompted them to launch "an effort to evaluate it on all open Millennium Prize problems and a few other high-impact problems."

It's not obvious to me that's an unethical thing to do, if it happened as they described.

Mark Litwintschik's blog is one of the best engineering blogs on the internet. Always practical, with detailed write-ups that make it easy to reproduce workflows. Thanks, Mark.

The Winter 2016 edition of the Center for Land Use Interpretation's newsletter, "The Lay of the Land", has more information on the cards: https://clui.org/newsletter/winter-2016/los-alamos-rolodex

There's also a book: https://clui-shop.square.site/product/los-alamos-rolodex/292...

I saw the exhibit in person back when it was on display. If you're ever in LA you should try to stop in to the CLUI's Culver City location, they always have something interesting.

Earlier discussion of the University of Chicago Law School's policy: https://news.ycombinator.com/item?id=49024980

In this case it was an own-goal. If the U.S. military hadn't jammed this flight's GPS it wouldn't have been jammed.

Worldwide, it's an issue that is going to require changes to the infrastructure of aviation: Training, procedures, equipment. That's what I mean when I say "they should have just used VORs and ILS" shows an unsophisticated understanding of aviation safety. "Just use the training you're already supposed to have" isn't enough. We need to make sure that crew and ATC aren't just improvising when there are GPS outages. Manufacturers need to ensure that equipment has been thoroughly tested against GNSS jamming and spoofing -- The Embraer Phenom used to start "UNEXPECTED ROLLING AND YAWING OSCILLATIONS (DUTCH ROLL) AT HIGH AIRSPEEDS,"[1] after it lost GPS; recently there have been many reports of inertial navigation systems becoming "contaminated" after GPS interference such that they give completely wrong positions for the rest of the flight, even after GPS returns.

You're right that just stopping the U.S. military from jamming civilian aircraft GPS doesn't solve the problem for the rest of the globe. That will take more effort, and it's why aviation authorities and industry have been looking at the problem for years. But it would mostly solve the problem for now, in the U.S., where significant GPS interference is rare outside of military exercises.

1. https://www.faa.gov/documentlibrary/media/notice/genot_7110_...

It's not interesting: they were not in restricted airspace at any point during the flight. If they were, probably the NTSB, a YouTuber, or I would have mentioned it.

There are sooo many GPS interference exercises NOTAMed! Here's an animation I made of GPS interference NOTAMS in effect on any given day in the U.S. from 2022 to 2023: https://x.com/lemonodor/status/1722275349554487750

And they affect thousands of aircraft--Here's a thread where I analyze an earlier exercise at White Sands, called NAVFEST: https://bsky.app/profile/lemonodor.bsky.social/post/3ml7afb2...

  * Approximately 4000 aircraft experienced significant GPS interference
  * 90% of which were commercial jets
  * The military appears to have been jamming outside the posted NOTAM hours
    2750 aircraft jammed during the window, 1250 jammed outside the window.
  * Used to be called JAMFEST but "the Joint Spectrum Center became
    uncomfortable with the idea of associating its RFAs with events with the
    word 'jam' in their names."

I run GPSJAM.org, have been studying and tracking the effects of GPS interference on aviation for the past 5 years, and I was a source for this article. I'm not a pilot and I don't work for any aviation agency.

I'm very curious to see what the final NTSB report says, but their preliminary report along with commentary from other analysts seems to show that this was a crew that made bad choices and died because of it. It also seems to show that GPS interference was a contributing factor to the accident, and that they'd likely be alive if the U.S. military hadn't been jamming GPS.

I think pilots need to be prepared to fly without GNSS, and also it might be a bad idea for the military to regularly deny GPS to thousands of civilian aircraft with tens or hundreds of thousands of passengers.

Lack of GPS isn't a critical safety issue, but it is a safety issue. GPS interference removes options, hurts situational awareness, disables or degrades other safety equipment (like TAWS, the Terrain Awareness and Warning System that is specifically designed to warn pilots that they're about to hit a mountain), and adds to crew and ATC workload and distraction.

Saying that this situation is what pilots train for and they should have just used VORs and ILS reveals an overly macho, unsophisticated understanding of aviation safety. We've added GPS, ADS-B, TCAS, GPWS, TAWS, etc. because they make flight safer. The airline industry and governments are concerned about GPS interference because of the aviation safety issues--if you're not concerned, you're thinking about the problem in a very narrow way ("I would never have trouble if my GPS died.").

In fact they think it's an urgent problem.

2024: "Aviation sector seeks urgent solutions for GPS interference" https://www.reuters.com/business/aerospace-defense/aviation-...

From a 2023 presentation by Eurocontrol at the UN International Committee on GNSS/Interference Detection and Mitigation https://rntfnd.org/wp-content/uploads/Aviation-GNSS-interfer...:

  • GPS Problem reports dominate over all other type of safety reports
  • Redundant systems are the only reason why aviation has been able to
    maintain normal operations despite GNSS RFI!
  • GNSS integrated into many systems; exact RFI impact difficult to
    predict, manufacturers had to issue aircraft specific
    guidance → Complexity and workload increase

  Aviation Safety Impact
  • Aviation Safety is built on two main principles:
  • Trust your instruments
  • Follow standard operating procedure
  • GNSS RFI causes pilots to have to question both principles!
  • Chief Operations Officer of one major airline:
    Navigation is not my problem. My problem is “normalization of
    deviance”!
  • Incidents have occurred simply due to pilot distraction because of
    having to deal with too many system alerts

The U.S. military jamming GPS of civilian aircraft is a situation where there have been several close calls before this incident and experts have been saying for years that if the military keeps doing this, even with NOTAMs, people will get hurt and then that prediction tragically came true.

From 2021, "FAA Files Reveal a Surprising Threat to Airline Safety: The U.S. Military GPS Tests" https://spectrum.ieee.org/faa-files-reveal-a-surprising-thre...:

  Early one morning last May, a commercial airliner was approaching El
  Paso International Airport, in West Texas, when a warning popped up in
  the cockpit: “GPS Position Lost.” The pilot contacted the airline’s
  operations center and received a report that the U.S. Army’s White
  Sands Missile Range, in South Central New Mexico, was disrupting the
  GPS signal. “We knew then that it was not an aircraft GPS fault,” the
  pilot wrote later.

  The pilot missed an approach on one runway due to high winds, then
  came around to try again. “We were forced to Runway 04 with a predawn
  landing with no access to [an instrument landing] with vertical
  guidance,” the pilot wrote. “Runway 04…has a high CFIT threat due to
  the climbing terrain in the local area.”

Also

  This is far from the most worrying ASRS report involving GPS jamming.
  In August 2018, a passenger aircraft in Idaho, flying in smoky
  conditions, reportedly suffered GPS interference from military tests
  and was saved from crashing into a mountain only by the last-minute
  intervention of an air traffic controller. “Loss of life can happen
  because air traffic control and a flight crew believe their equipment
  are working as intended, but are in fact leading them into the side of
  the mountain,” wrote the controller. “Had [we] not noticed, that
 flight crew and the passengers would be dead. I have no doubt.”

Here's what an air traffic controller with 18 years of experience wrote about WSMR GPS interference in 2024: "Someone will get hurt ignoring the pilot reports or deciding for pilots how much equipment can fail or be unreliable before they 'agree' or decide to stop GPS jamming." https://asrs.arc.nasa.gov/docs/rpsts/ctlr.pdf

A few months ago I built a piloting harness for LLMs[1] in X-Plane. You can give the AI pilot instructions as though you are ATC, and I imagined what it might be like to scale it up and make an ATC game out of it, talking to dozens of AI pilots. The most open-ended ATC sim ever…

[1] https://www.youtube.com/watch?v=UiXcLURL7yo

Mr. Show, "Change for a Dollar": https://www.youtube.com/watch?v=KyocQT4Vn2g

This is more useful than every other "monitoring the situation" dashboard I've seen.

The Aisle "the moat is the system, not the model" blog post comparing Mythos' results to their system's was misleading, and seemed to be an attempt to ride the coattails of attention on Mythos. It was of low enough quality that I'd want to see more details of exactly how these vulnerabilities were found.

And in Bartz v. Anthropic, the court found that Anthropic training their LLMs on books was "highly transformative."

Thanks for the details and correction.

From what I see the code downsamples video to 5 fps, so 1 hour of video is 3600 seconds * 5 fps = 18,000 frames. 18,000 frames * $0.00079/frame = $14.22. A couple dollars more with the overlap.

(The code also tries to skip "still" frames, but if your video is dynamic you're looking at the cost above.)

Jamming is a good way to make sure everyone knows exactly where you are.

I find it odd that there's a custom of blurring or obscuring exactly the thing I'm interested in when I show interest in it by mousing over it.

Since this post is a dupe, here's a video demonstrating a similar but different app I made: https://www.youtube.com/watch?v=EjH0kMEz4YY

This went from #2 on the front page to the bottom of page 2 very quickly. It's unfortunate.

Or one person types 76 pages. This is a thing people used to do, not all that infrequently. Or maybe you have one friend who will help–cool, you just cut the time in half.

You're just saying things that don't have basis in reality.

It's not something halfway critical to the operation–why would the FAA allow that? ADS-B Exchange is not run by volunteers–it's run by employees of JETNET LLC, an aviation intelligence company. Doing it yourself almost certainly gives you less information–you're not part of a global network of receivers. It almost certainly gives you less reliability–receivers in the big networks typically have a fair amount of overlap which gives redundancy your single receiver doesn't have.

It's also not FAA approved!

Why? You would almost certainly get better data with higher reliability and no effort and no money spent from airplanes.live, adsbexchange.com, etc.

It's happening again. Spoofing is in progress, rendering another image. ADS-B Exchange has blocked access to the ICAOs/hexes in question--if you try to look at their history you get redirected to the base map.

https://x.com/TheIntelFrog/status/2016841289556168990

airplanes.live, adsb.lol, ADS-B Exchange, adsb.fi, etc. do not censor the data.

These receivers mostly don’t have gps and it’s very common for people to put in the wrong coordinates.

A transponder in a car is not an "aircraft station" (§ 87.5), therefore it is not covered by aircraft "license-by-rule" (§ 87.18(b)), so transmitting would be operating without a valid authorization (§ 1.903(a)). https://www.ecfr.gov/current/title-47/chapter-I/subchapter-D...

Imagine your boss doesn’t like you looking at ADS-B sites because it’s not data from an FAA approved system but as long as you’re discreet and not actually breaking a reg they don’t yell at you. Then they come in and see that you installed an antenna, RTL-SDR, and raspberry pi in the tower.

Just because I don't often get a chance to talk about this, I'll mention that there was a malfunction/accident/bug that caused what you might call spoofed signals to go out around Long Island and New York. Really interesting case where it seems that an FAA system wasn't handling magnetic declination correctly, which led to it generating false TIS-B targets that were rotated 13 degrees from real aircraft positions, from the radar antenna point of view: https://x.com/lemonodor/status/1508505542423064578

(TIS-B is a system that broadcasts ADS-B-like signals for aircraft that are being tracked by radar but either don't have ADS-B Out or otherwise might not be picked up by other aircraft with ADS-B In, e.g. maybe they're at a low altitude.)

There have been a couple other incidents with the TIS-B system. E.g. this apparent test near Dallas in 2022 that generated dozens of false targets in an interesting pattern: https://x.com/lemonodor/status/1481712428932997122 There was a similar incident around LAX several months later.

There are non-radar towers that don't have scopes. They may have a traffic display, or maybe not. They might choose to use a public ADS-B aggregator site because it gives them situational awareness, but they don't use it to provide radar services to aircraft. That's my understanding from listening to a lot podcast episodes with air traffic controllers, anyway. I think it's an unofficial, non-FAA approved kind of thing that can make their jobs easier.

See https://www.faa.gov/air_traffic/publications/atpubs/atc_html... for non-radar ATC procedures.

That's not true. And if you click almost anywhere else on the spoofed track it will show as Source: ADS-B.