Blog post #94

AI-generated illustration.

---

The Prize in Economic Sciences was the sixth and last Nobel announcement of the week, at 11:45 on Monday. It went to Robert E. Hall and Michael Woodford “for microeconomic foundations of macroeconomic stabilisation”. The film went live a little after one in the morning, which wraps up the whole series: six prizes, six films.

🎥 Prefer video? Watch the film on YouTube.

---

What shipped

• A 15:50 film on the channel Superintelligence, Explored, with an English main track and a Swedish audio track, chapters, sources and image credits. It is in the playlist Nobel Prize Week 2026, Explained together with medicine, physics, chemistry, literature and peace.
• The story: why the economy goes up and down, why the old Keynesian tools broke in the 1970s oil crises, Hall’s test of how households spend, Woodford’s three equations, why central banks steer with their words (forward guidance), and how the models handled 2008, the pandemic and this year’s oil shock around the Strait of Hormuz.
• A scene with real math. I missed the equations in the first cut, so the second version got a new scene where the three equations of the simplest New Keynesian model are shown on screen. Each term lights up when the voice explains it, and at the end every E (for “expected”) glows at once. That is the whole point of the prize: expectations sit in every equation.
• The previous film, on the peace prize, now links to this one in its description.

How the day went

• 11:45: the announcement. I watched the stream and the interviews and typed notes straight into the chat while the agent set up the project from my Nobel kit and pulled the press release and both background PDFs from nobelprize.org.
• Research and script: a research agent wrote a source-tagged research file while the first script was written from the popular background. My interview notes went in as I typed them: the house-buying example, the point that the 1930s got stuck, and the committee saying the model works like a laboratory.
• Four reviewers: Codex reviewed the script against the Nobel documents before any voice was recorded, then Grok and ChatGPT reviewed it too. They disagreed once in a useful way: one of them said AP quoted Woodford on explaining policy to the public. We opened the AP article and the quote was not there, so that line now quotes the Academy’s press release instead.
• Voice, images, build: ElevenLabs eleven_v4 with my voice clone, 22 images generated with Codex, six public domain photos, and four agents building the scenes in HyperFrames in parallel.
• Three rounds of feedback. v1 was fine for a first look, but many animations were sloppy: dancing green blobs for people, a wobbly speedometer, clumsy hand-drawn icons. v2 got photo-style images, miniature houses and shops, real Lucide icons and the equation scene. v3 fixed the last details: a clearer pandemic marker on the roller coaster, an AI marker at the bottom of the curve where the economy turns up again, a price tag that flips from $100 to $120, and a real portrait of Kevin Warsh.

Decisions made

• The AI angle. The Nobel documents never mention AI, and the first AI scene felt forced. The angle I ended up with: Hall and Woodford wanted an economy built from millions of individual decisions, but in the 1990s you could only model one typical household. With AI you could simulate far more detail at the micro level and test policies on the simulation first. We are not there yet, but in one experiment (EconAgent) households run by language models produced the Phillips curve on their own.
• My own take: communicate more and be as transparent as possible, so households and businesses can adapt. That puts me close to the laureates and against the new Fed chair Kevin Warsh, who has pulled back from forward guidance this year. The film shows his argument fairly and does not bring Trump into it.
• No criticism of the prize. Some critics called it badly timed. Instead the film shows the committee’s own list of what the models still cannot do.
• Portraits. There is no freely licensed photo of Woodford anywhere, and the Reuters and Nobel portraits are not free to use on a monetised channel. Hall gets the CC0 portrait from Wikipedia, Woodford gets a name card. Warsh’s official Federal Reserve portrait is public domain, so he got a real photo.

What’s unclear or broken

• The voice sounds a bit echoey. My professional voice clone was trained on the older ElevenLabs models, and v4 training for it has not reached my account yet. When it does, I will regenerate the narration and re-render.
• Codex could not save generated images into the project folder because of its sandbox. The images were still in its own folder, so the script now copies them from there.

Tooling & process

• HyperFrames for the film, with four agents building scenes in parallel and a harness that checks every scene before render.
• ElevenLabs eleven_v4 with my voice clone for English and Swedish, Parakeet transcription to check that no words were dropped.
• Codex for images and fact checking, Grok and ChatGPT as extra reviewers, and the YouTube Data API for upload, chapters, localisation and playlist.

---

— Stefan