Full Width 
Shortcuts 
Sign Up 
Log In 
Articles 
Reading List More from John Wiseman
Two projects at the intersection of geospatial data and AI. First, a semantic search engine for aerial imagery, fine-tuned on OpenStreetMap-tagged data. Second, WARP—a custom neural model that learns embeddings of aircraft flight trajectories, trained on roughly a million aircraft tracks. Given any flight path, WARP can instantly retrieve the most similar trajectories from history. It enables applications from OSINT and anomaly detection to safety analysis and flight planning. The same latent structure that made word2vec transformative for language now becomes available for aircraft behavior. Presented at Spec.LA on December 18, 2025. I'm going to quickly show two things tonight. The past few years I've been working mostly in geospatial, aviation, and AI, and that's what these are. One is geospatial semantic search and the other is a foundation model for aircraft. I believe they're both extremely valuable and useful. # Both projects are about embeddings. Embeddings are points inside a latent or conceptual space. Parts of that space that are conceptually similar are closer together. There are neural models that will generate an embedding of an image or text or other things. With the right model, an embedding of a photo of a golden retriever is close to the embedding of the text "golden retriever" and far from the embedding of an image of an aircraft boneyard. # I showed an early version of PIMINTO at Spec a long time ago. I've refined it a lot since then. It uses a model that's fine tuned on satellite imagery and text describing the imagery based on OpenStreetMap tags. I'm going to give a live demo of the new site, which is public. piminto.obliscence.com | Demo video # Next is WARP, which is a model I'm building to create embeddings of aircraft paths. Back to aircraft, big surprise. # I trained the model on about a million aircraft tracks, pretty much everything that flew in the western U.S. over the course of a week. # At this point training takes 18 hours on a 4090 GPU. # I've come up with a custom architecture for this task. # I kept feeding the single file into GPT o3 or GPT 5, either asking it to critique the code or describing specific problems I was running into. Sometimes I'd paste a CSV of the loss curve. It would give me detailed descriptions of what to do next, and I'd paste those into Cline with Sonnet 4.5 doing the actual coding. # How does it do? # I hoped that if I took a few thousand trajectories and clustered their embeddings I would see clear separation and similarities. But that didn't really happen. A few extreme examples looked good, like skydiving aircraft being outliers, but otherwise I couldn't really make sense of it. But then I read someone who said sometimes clustering just looks nonsensical even if the embeddings are good at capturing similarity. # So I wrote an interface to let me compare any of the million trajectories. I can select one trajectory and it will almost instantly show me the other most similar trajectories in the database. In this screenshot the blue is the query trajectory and the red is the most similar trajectory it could find. This one isn't too bad. The altitude and speed profile are very similar but not exactly the same. Promising. # This is the signature of touch and go practice. This is a good match. # These are two different skydiving aircraft. They climb up to altitude, everyone jumps out, then they race back to the field. # These are two different flights by the same aircraft. But what does it really look like? # This is pretty cool. This looks like two training flights, by the same aircraft but on different days. Each time following the same route and including kind of a loop and then some tight circles. # These are two different Thales Watchkeeper drones, flying on different days over Fort Bliss. As a reminder, I chose the track on the left and then the system searched the database in a fraction of a second and determined that the track on the right was most similar to it. # This starts to hint at some potential uses. Both of these helicopters are doing some sort of utility work in forest. Hauling something maybe. Again, the track on the left is the query track, and the one on the right is the best match. # These are two different aircraft on two different days, dropping sterile fruit flies on different parts of Los Angeles. There are many other aircraft flying these sorts of lawnmower patterns, for example doing LIDAR collection, but it determined the one on the right is the most similar. # A useful aircraft embedding model unlocks a lot of interesting and valuable use cases, similar to how word2vec and text embeddings unlocked a whole class of semantic search, clustering, retrieval, anomaly detection, and analogy-style reasoning over language—except now the same kind of latent structure becomes available for aircraft behavior, trajectories, and operational patterns. # # { img.style.cursor = 'pointer'; img.addEventListener('click', () => { lightboxImg.src = img.src; lightbox.classList.add('active'); }); }); function closeLightbox() { lightbox.classList.remove('active'); } document.addEventListener('keydown', (e) => { if (e.key === 'Escape') closeLightbox(); });
More in technology
A Pika author was editing the same post in two browsers. They added six images in Browser A, then later they referenced an old tab editing the post in Browser B. That copy of the editor did not include those images. Browser B saved and removed all six images from Active Storage. A short time later, Browser A saved its newer copy of the post and put the image references back, but at that point the originally uploaded files were gone due to Browser B saving the post without the images. This exposed two separate problems: Missing images produced uncached 404 responses that the editor requested repeatedly. Action Text purged an image as soon as one save removed it from the body. Here’s how we fixed those problems. Why Action Text deleted the images Action Text keeps a rich text record’s embeds synchronized with the attachments in its body. When a submitted body no longer contains an image, Action Text removes its ActiveStorage::Attachment. Active Storage then purges the dependent blob, deleting both its database record and its file. That behavior makes sense when there is only one current copy of a document. It is unsafe when the same post can be open in multiple tabs, browsers, or devices. One stale save can delete an image that another editor is still using. Why the missing images flooded the server Browser A still had the six images in its editor, each pointing at an Active Storage redirect URL: /rails/active_storage/blobs/redirect/:signed_id/:filename Pika uses the ProseMirror-based Rhino editor. Rhino rebuilt an image’s DOM node when its caption or selection changed. Each new <img> element requested its URL again. Healthy images did not cause trouble since Active Storage makes its redirect cacheable. For a purged blob, Rails returned a 404 without Cache-Control, so the browser kept sending the request to Rails instead of reusing the failure. In a gallery, typing could re-request several missing images per keystroke. In these scenarios, this issue led to hundreds of requests hitting our servers. Fix 1: Cache permanent Active Storage 404s A purged blob’s redirect URL can never start working again. Blob IDs are not reused, and uploading the image again creates a new blob with a new signed URL. Sinc there is no concern of a 404 becoming stale, we made a specific 404 cacheable for the same period as a successful redirect: # app/controllers/concerns/active_storage_missing_blob_cache.rb module ActiveStorageMissingBlobCache extend ActiveSupport::Concern included do rescue_from ActiveRecord::RecordNotFound, with: :render_cacheable_not_found end private def render_cacheable_not_found expires_in ActiveStorage.service_urls_expire_in render file: Rails.public_path.join("404.html"), layout: false, status: :not_found end end We included the concern only in the controller serving these blob redirects: # config/initializers/active_storage_missing_blob_cache.rb Rails.application.config.to_prepare do ActiveStorage::Blobs::RedirectController.include(ActiveStorageMissingBlobCache) end (We also offered an update to Rhino that avoids it being so aggressive re-requesting images.) Fix 2: Detach dropped images instead of purging them When a save removes an image from an Action Text body, we now remove the attachment while temporarily keeping its blob. Another browser can save the image again and reattach that same blob. If nothing brings it back, a cleanup job purges it later. First, we added a timestamp for when the image was dropped: class AddDetachedAtToActiveStorageBlobs < ActiveRecord::Migration[8.1] def change add_column :active_storage_blobs, :detached_at, :datetime end end We then intercepted the private callback Active Storage uses to purge a dependent blob: # config/initializers/detach_dropped_embeds.rb module DetachDroppedEmbeds private def purge_dependent_blob if dropped_from_rich_text? blob&.update_columns(detached_at: Time.now) else super end end def dropped_from_rich_text? record.is_a?(ActionText::RichText) && !destroyed_by_association end end ActiveSupport.on_load(:active_storage_attachment) do prepend DetachDroppedEmbeds end The destroyed_by_association check preserves intentional deletion. Removing an image from a saved body detaches it, but deleting the post and its rich text still purges its images normally. Purge detached blobs after a grace period The cleanup query considers both when the blob was created and when it was detached: ActiveSupport.on_load(:active_storage_blob) do scope :unattached_since, ->(time) { unattached.where( created_at: ..time, detached_at: [nil, ..time] ) } end That scope handles both kinds of unattached blobs: A new upload that has never been attached becomes eligible based on created_at. An image removed from an old post becomes eligible based on detached_at. Pika gives both a 30-day grace period: class PurgeUnattachedUploadsJob < ApplicationJob GRACE_PERIOD = 30.days def perform ActiveStorage::Blob.unattached_since(GRACE_PERIOD.ago).find_each do |blob| blob.purge_later end end end If another browser saves the image during that period, it is attached again and the unattached scope excludes it from cleanup. The result Replaying the original two-browser edit now works safely: Browser B saves a body without the six images. The attachments are removed, but their blobs are only marked as detached. Browser A saves its newer body and reattaches the existing blobs. The images remain available and do not need to be uploaded again. If no editor brings a detached image back, the cleanup job eventually purges it. Much better! Reply by email
Yesterday, a group of MIT researchers released an article titled “Sparks of Weight-Space Programming: LLMs can natively read & write neural network weights”. The piece describes a set of experiments which convincingly show OpenAI’s latest Astra model is able to understand and successfully alter raw weight data. Wowzers! It’s a rough analogy, but an LLM’s weights are sort of like its brain. A set of weights is a large pile of numbers generated during training, and an LLM’s output is derived by running input through the weights using some matrix multiplication and a bit of randomization. While the steps to generate weights are well understood, the weights themselves are extremely opaque. There’s ongoing work to try and explain how weights work, but researchers currently have a limited understanding of what these numbers do. Based on this, I was quite surprised to learn that LLMs are able to act on raw weight data — even if it’s still just in limited cases. It’s unclear if OpenAI’s model was trained on weight data. Even if it was, I still find these results surprising. The article shows the model accurately predicting what a never-before-seen set of weights was built to do, fixing another model by updating its weights, and creating new sets of weights that performed simple tasks. I would not have guessed any of that would be possible a couple of days ago. This whole thing is slightly unnerving. The whole “weights as a black box” thing has always felt a bit sketchy to me, even when I was working with much simpler machine learning models. Now these experiments hint at a future where LLMs could be self-improving in ways that aren’t human-verifiable. Machines building machines, and we can only hope that they’re not doing anything weird. I don’t know if this strategy is scalable, but I’d guess that it’s something more people will be exploring very soon. More and more often these days, I wish we were living in less interesting times.
Harry threw me off course in episode 48 of Traction Heroes. At first, I thought was reading from a book about mind mapping. Soon, it became clear it was about something else: how to define a clear vision for a team or organization. Visions are notoriously hard to articulate. Drawing a picture — especially a picture of the whole forest, plus the trees — can help. The passage Harry read from Cameron Herold’s Vivid Vision explores this idea: So to visualize something a little more tangible, think about a tree trunk in the earth, inner core with its branches stretching out into the sky. Each of those branches contains twigs extend further out. The mind map works in a similar way, a core that branches out into smaller components. Now, apply this to the world of business. As the leader of an organization, you’re firmly planted at the center, and the various areas that contribute to the business branch out: marketing, sales, engineering, and so forth. With that in place, build off that by describing those branches in about three or four bullet points each. In other words, describe what your marketing looks like and sales and so on until you’ve captured every idea. Remember too that among the branches of your business, you want to consider every aspect of the organization: the vibe, informal codes of conduct, the fiscal environment, the scenery, all of that. Deliberately leave out numbers and data. Get a feel for the place. When you step out of your time machine three years from now and walk around the office, what’s happening? What can you say that is typical of the business three years from now? The mind map is an initial foray into the vivid vision and will act as an outline for the rough draft that you, the leader, will then write. I’ve sketched diagrams before when outlining articles, workshops, and presentations, never in service of aligning around a company’s vision. Harry’s done it several times and found it easier than trying to write it out. Seems useful — check out our conversation for more: Traction Heroes episode 48: Visions
Sharing a few data points because some people did not expect these drives to last that long.
I'm 100% for banning social media for children. I would even extend it to banning it all together in its current form. But I am also against the methods for enforcement. The problem with banning social media for one group and not another is that you have to differentiate the different groups somehow. And right now, in an age where information and personal data means everything, there are several entities that are salivating to get this data. I've written before about all the problems that come from uploading sensitive data online. It doesn't even matter if they delete the data afterward, as long as the data goes through the wire in the first place, there is a chance for a leak. What appears to be the same simple request "please verify your identity", becomes fundamentally different when mediated by technology. The question isn't whether these digital systems will be compromised, but when. And unlike that movie theater clerk who can't perfectly recall my birthdate minutes after seeing it, computers have perfect memory. They store, copy, backup, and transmit our most sensitive information through networks we don't control, to companies we've never heard of, under policies we'll never read. When kids are not allowed to use social media, it's the adult who pays the price. It's the adult who has to prove that they are not children. They have to risk uploading their government ID to a third party, or biometric data that may be saved and backed up forever. I reject this method. I also reject the method where I have to provide my age to the operating system. No system should compel me to provide personal data in order to use it. Especially if this feature is intended to provide the information to a third party. And what happens if I don't provide this information to my freebsd machine? Or worse, what if I provide false information? Am I going to get punished? How does that protect children in any way? Breaking news: Grown up arrested when OS reported forged identity in his laptop. One question I asked myself was what happens when a minor circumvents the ban. What are the consequences? At least in Australia, a country that has implemented a social media ban for those under 16, there are no consequences. The law places legal responsibility on platforms, not children or parents, and it currently applies to Facebook, Instagram, Kick, Reddit, Snapchat, Threads, TikTok, Twitch, X and YouTube. This almost sounds reasonable. Why should you care, if it becomes facebook responsibility when you don't even own a facebook account? But there is always a caveat. When facebook becomes a tool your school uses to communicate with parents. When Google classroom is the service students use to do their homework. When you are required to login on a system to receive government service. You may not care about a particular social media, but they raise the bar for every other entrant in the game. Every website frequented by a large number of people inherently become social media, and demand data you didn't need to provide before. Technology is not a magical thing that can just solve every problem. It is especially weak when it comes to social problems. However, society can use technology to solve their problems. The responsibility of children falls onto their parents. And there are great tools for parents that their kids access on their devices.
