Zero-shot forecasting and the nature of time

Did the Odyssey really happen? The truth of poetry lies in weather

Now, goddess Child of Zeus tell the old story for our modern times. Find the beginning… We are once again looking for the beginning, but not of a model. Instead, we're looking at ancient weather observations in Homer's Odyssey — perhaps one of the greatest weather texts ever written. We examine a paper that reconstructs the storm that shipwrecked Odysseus, exploring how modeling past weather relates to future forecasting and what it means to validate a 2,800-year-old poem as archival meteorological observation. ---------------------------------------- PAPER * Meteorological Assessment of Homer's Odyssey [https://journals.ametsoc.org/view/journals/bams/74/6/1520-0477_1993_074_1025_maoh_2_0_co_2.xml], Cerveny, 1993 ---------------------------------------- CHAPTERS * 00:05:22 - A lil refresher of what went down in the book * 00:13:16 - Paper time! Intro to reconstructing ancient weather * 00:20:09 - Day-by-day on the Mediterranean sea * 00:36:45 - Actually NOT reconstructing the weather, but validating the Odyssey's accuracy * 00:41:52 - Why you, weather person, should read the classics ---------------------------------------- RECOMMENDED READING * The Odyssey by Homer (Translated by Emily Wilson) * Works & Days by Hesiod (Translated by A. E. Stallings) * The Three-Body Problem by Liu Cixin

Zero-shot forecasting and the nature of time

In this episode, we're covering two papers on zero-shot forecasting: NXAI's TiRex and Amazon's Chronos-2. You may be asking… Is it pronounced tee·rɛks? Is it tye·rɛks? Is it a titan? God of time? Is a time series just a sequence? Is a sequence just a sentence? Is time a sentence? Is time a poem? As a poem constellates images, and an LLM strings together tokens, the authors apply this approach to time series forecasting, offering new opportunities for zero-shot weather prediction. We discuss the history of the term "Zero-shot," breakdown each paper from training data to industry applications, and wax poetic about the paradigm shift these models are bringing to earth systems forecasting. ---------------------------------------- PAPERS * TiRex: Zero-Shot Forecasting Across Long and Short Horizons with Enhanced In-Context Learning [https://arxiv.org/abs/2502.00479], Auer et al * Chronos-2: From Univariate to Universal Forecasting [https://arxiv.org/abs/2410.04220], Ansari et al ---------------------------------------- RECOMMENDED READING * The Crying of Lot 49 by Thomas Pynchon * God, Human, Animal, Machine: Technology, Metaphor, and the Search for Meaning by Meghan O'Gieblyn * The Odyssey by Homer (Translated by Emily Wilson) * The Hainish Cycle series by Ursula K. Le Guin

A taxonomy of bias: sensemaking, heretical physics, and the Tom Hanks/Bill Murray multiverse

Bias has a pretty well-known definition in the world of AI/ML programming. But today’s paper asks us to expand that definition and consider how cultural, organizational, and human forces can intersect with development. The authors of “Identifying and Categorizing Bias in AI/ML for Earth Sciences” argue that, beyond considering bias and before mitigating bias, developers ought to be able to identify bias in all its modes & forms. Their taxonomy of bias is helpful, and inspired us to create an actionable check list of our own. ---------------------------------------- PAPER Identifying and Categorizing Bias in AI/ML for Earth Sciences [https://journals.ametsoc.org/view/journals/bams/105/3/BAMS-D-23-0196.1.xml], McGovern et al ---------------------------------------- CHAPTERS * 00:00:03 - Intro * 00:02:39 - Abstract * 00:03:33 - Discussion of framing * 00:09:03 - A taxonomy of bias 09:03 * 00:47:52 - Our proposed check list for mitigating bias * 01:08:14 - Reading recs ---------------------------------------- A check list for mitigating bias (in developing AI or for really any kind of technological development) * Use the language of “bias”, acknowledging both social and technical bias * Define your specific user and use case early * Establish a clear baseline for comparison * Ensure diversity of perspectives on your team * Practice reflexivity (question your assumptions, and continue that line of questioning throughout the development process.) * Examine your incentives vs. end-user incentives * Prioritize transparency (in your decision-making and in your incentives & goals.) Recommended reading * AI2ES Newsletter [https://backup.ai2es.org/] * Groundhog Day (film AND musical theater adaptation) * “Science as a Vocation [https://sociology.sas.upenn.edu/sites/default/files/Weber-Science-as-a-Vocation.pdf]” by Max Weber * Book of the New Sun by Gene Wolfe (four-book series) * Infinite Powers by Steven Strogatz

Determinism is dead, chaos reigns, and the night is long

While staring into uncertainty might sound abyssal and frightening, ensemble models have proven (to us, at least) that this isn’t the case. In today’s episode, we’re exploring two papers with different approaches to ensemble forecasting. This "choir" approach to weather prediction is one that embraces chaos rather than striving to chart a single line of truth on a graph by generating dozens or even hundreds of slightly different predictions that together map a full range of possible outcomes. We’re unpacking FGN, the latest variant of AIFS, and their differing approaches to the same challenge: how can we create a confident forecast of an assuredly uncertain future. ---------------------------------------- FEATURED PAPERS * Skillful joint probabilistic weather forecasting from marginals [https://arxiv.org/abs/2506.10772], Alet et al. * AIFS-CRPS: Ensemble forecasting using a model trained with a loss function based on the Continuous Ranked Probability Score [https://arxiv.org/html/2412.15832v1], Lang et al. ---------------------------------------- CHAPTERS * 00:00 — Intro * 01:38 — Abstract (Meet the paper!) * 10:17 — Weather report & reading list * 19:37 — A lil history (Newton, Blake, Goethe, & voices of dissent) * 28:43 — Paper time! ---------------------------------------- FURTHER READING * The Solace of Open Spaces by Gretel Ehrlich (especially “On Water”) * North Woods by Daniel Mason * Frederick by Leo Lionni * Hyperion and The Fall of Hyperion by Dan Simmons * Newton [https://www.tate.org.uk/art/artworks/blake-newton-n05058] by William Blake * But how do AI images and videos actually work? [https://www.youtube.com/watch?v=iv-5mZ_9CPY] by WelchLabs and 3Blue1Brown (video) * Chaos: Making a New Science by James Gleick * “From chaos to clarity: Seasonal forecasts for confident risk management” [https://www.upstream.tech/posts/from-chaos-to-clarity-seasonal-forecasts-for-confident-risk-management] with Marshall, Alden, and Phil Butcher from the HydroForecast team

NeuralGCM and the Hybrid Approach

Machine learning dominates the conversation, but what will happen to the centuries-old physical equations that built our understanding of the atmosphere? How are the two approaches at odds? How might they coexist? Today’s paper, NeuralGCM, sheds light on how physics-based and AI approaches might be a powerful pairing for weather forecasting. For the first truly hybrid model we’ve discussed, it’s only fitting that we’ve also taken a hybrid approach in this conversation. So join us for hybrid models, data compression, Dragon Ball Z, and the strange future of primitive equations. ---------------------------------------- FEATURED PAPER Neural general circulation models for weather and climate [https://www.nature.com/articles/s41586-024-07744-y] ---------------------------------------- CHAPTERS * 00:00 Intro * 01:31 Weather report & books * 18:01 Paper time * 25:26 A theory of compression * 48:07 Closing thoughts: Resolution, interpretability, & the future of primitive equations ---------------------------------------- RECOMMENDED READING * Chaos: Making a New Science — James Gleick * Landmarks — Robert McFarlane * Dandelion Wine — Ray Bradbury * Webster’s 1913 Dictionary [https://www.google.com/url?q=https://www.websters1913.com/words/Wind] * “The Bitter Lesson” [http://www.incompleteideas.net/IncIdeas/BitterLesson.html] — Rich Sutton * MC-LSTM: Mass-Conserving LSTM [https://arxiv.org/pdf/2101.05186] * Charisma and Disenchantment — Max Weber