Jordan maps Los Angeles parking tickets from open city data with Mapbox and Codex, then we test Sora 2 cameos and compare its video quality with Veo 3.
You can map Los Angeles parking tickets yourself because the city publishes its parking citation data through downloads, queries and APIs, though it runs about two days behind, so it can't show where officers are right now. Jordan built that map on Episode 15 of Built This Week with Mapbox, Cursor and OpenAI's Codex. Then Jordan and Sam tested Sora 2 on its first day out, including the cameo feature that puts your own face in the videos.
Jordan saw a post on Twitter from someone who had gained access to an API showing where parking tickets were being issued in San Francisco. The map showed in real time where officers were writing tickets. It went a little viral, and Jordan says the city took the parking system down multiple times to block access, defending the move as protecting officer safety. Residents loved it, because they could see where officers weren't.
Parking tickets are one of those things that, we've decided as a society that we're okay with. But, everyone in society kinda hates them.
— Jordan Metzner, Episode 15
Watch Jordan map LA parking tickets in Episode 15 (starts at 3:10) · Episode 15 notes and transcript
Jordan had worked with Los Angeles city data before, including parking data at LAX. LA makes its parking data available through Excel, queries and APIs, so he queried the API and plotted the tickets on a map with Mapbox. He coded almost all of it in Cursor and OpenAI's Codex.
The app shows where tickets are issued, filters by individual officer, and has an officer leaderboard plus breakdowns by vehicle color, violation type, agency and date range. Some patterns jumped out. Beverly Hills and Santa Monica show no tickets, which makes sense because they are separate cities. In the Valley, tickets cluster along Ventura Boulevard, mostly for expired meters, and there is a long stretch between Sherman Oaks and Encino with very few.
On car color, black, white and gray cars get the most tickets, but they are also the most common colors. Brown, purple and gold cars rarely get tickets, though Jordan points out they are uncommon too, so he isn't recommending a brown car.
No. Sam asked whether the LA version could work like San Francisco's live map. Jordan found the data runs about two days behind, because officers write tickets on the street and upload them to the system later. He even found tickets dated in the future, which turned out to be mislabeled years, not real-time uploads. He thinks the delay probably makes it safer for officers too.
Sam's takeaway was that you can do a lot with available APIs to show data and make better decisions quickly. Jordan sees it as part of something bigger:
I think this is just like the beginning of, giving, residents and citizens access to their city's data to be able to look at it in different ways.
— Jordan Metzner, Episode 15
He also questioned whether ticketing is worth it for the city. The amounts in his dataset looked small for a city the size of Los Angeles, and he says he read a few months ago that the parking division was cash flow negative, though he did not verify that on the show.
Sora 2 came out the day before we recorded, and we both spent the day on it. Sam was most impressed by the fidelity of the face of whoever is generating the video, and by the social app OpenAI launched alongside it. After one day of use, though, he felt the physics in active clips, like skateboarding or playing soccer, were wonky compared to Veo 3. You can read more on our Sora vs Veo comparison.
Jordan says the physics are sometimes great, like a trampoline clip where you can see the surface flex, and sometimes off, like a hand with the wrong number of fingers. Generations took about three to five minutes. He had made 10 to 15 or more videos at no cost so far, compared with the $6 to $8 per video we paid when we were using Veo 3. Access was invite-only, and he was even turned away once because there weren't enough servers available.
Jordan says this is the first time we've seen this kind of character continuity. You film your face, moving your head left, right and around, and then you can put yourself inside any scene. He showed himself tanning leather, working on an oil rig, skydiving, riding an elephant and driving a Formula One car. Sam noticed the model often kept the black hoodie Jordan wore when he filmed his face.
There are limits. Sam says you can only make videos of other people if they have granted you access to their face and you are friends on the app, which is why so many videos feature Sam Altman. Jordan sees it as a new kind of social network.
Jordan's main takeaway after the first few days:
I think already I've noticed, the videos are interesting because you're the star of it and less interesting when you're not the star of it.
— Jordan Metzner, Episode 15
Clips are eight or nine seconds today, and he expects 20 to 40 second videos soon, with the same character carried through to tell a story. He expects problems with physics and continuity to fade until generated video is hard to tell apart from real footage. We came back to the Sora API and AI automations in Episode 17.
In the news segment, Jordan noted that OpenAI ran an employee share sale at a $500 billion valuation, above SpaceX. His explanation is that AI labs compete in three big verticals: chatbots, coding and content generation. A few weeks earlier he would have given coding to Anthropic and image and video to Google. In his view, OpenAI now leads all three, with GPT-5 for chat, Codex for coding and Sora 2 for video, though he had not yet spent much time with Anthropic's newest model.
Jordan walks through the LA parking map officer by officer and shows a reel of his Sora 2 cameos on camera. Watch Episode 15 of Built This Week, where the full transcript is also on the page.
