The Skeptics' Guide to the Universe · Dr. Steven Novella

The Skeptics Guide #1063 - Nov 22 2025

November 22, 2025

The episode opens with Steve Novella describing his trip to Dubai, where he delivered a nine-hour seminar on scientific skepticism as part of the FEEL fellowship program and participated in a two-day Future Forum conference. He notes Dubai is positioning itself as a city of the future, with heavy investment in technology and a corporate-flavored conference culture dominated by AI discussion. Steve and the hosts discuss the AI investment bubble, comparing speculative overvaluation to the dot-com era and 1929, and note that prominent investors have recently sold large NVIDIA positions. Evan Bernstein presents the Dumbest Thing of the Week: NBC's Today Show covering a 2025 American-Turkish research team claiming GPR scans of a Mount Ararat formation revealed manmade structures consistent with Noah's Ark. Evan explains that retired CSU geologist Lawrence Collins co-authored a paper concluding the site is a natural volcanic sediment formation shaped by erosion. Steve observes that ground-penetrating radar is becoming a favored pseudoscientific tool because its vague, low-resolution outputs can be interpreted as almost anything, analogous to ghost hunters using audio squelch machines. Jay Novella then covers a preprint from Stanford and the ARC Institute in which two California teams used an AI language model — from the EVO and EVO2 family, trained exclusively on DNA sequences — to design complete viral genomes for the bacteriophage PHIX174. Sixteen AI-generated designs produced functional phages that successfully infected and lysed bacterial cells in the lab. Jay explains bacteriophage biology, noting each phage typically targets only one bacterial strain, making them attractive for precision treatment of antibiotic-resistant infections. He cautions that the research is a preprint, not yet peer-reviewed or replicated, and that medical applications are likely decades away. The hosts discuss the dual-use risk: the same platform could theoretically be used to engineer bioweapons, and regulators are characteristically far behind. Steve recounts a Dubai conference panelist openly preferring to operate before regulators become aware of their work. Cara Santa Maria's pre-recorded segment describes a paper in the physics archive proposing a digital twin of the full Earth system at 1.25-kilometer resolution, using the icosahedral non-hydrostatic model developed by the German Weather Service and Max Planck Institute for Meteorology. The simulation covers both fast systems — weather, energy, and water cycles — and slow systems like the carbon cycle and ocean geochemistry, modeled in parallel using the CPU and GPU components of NVIDIA GH200 Grace Hopper superchips respectively. The model uses 672 million calculated cells representing roughly one trillion degrees of freedom. Running on 20,480 GH200 chips, it can simulate 145.7 Earth-days in a single real day. Cara notes the GH200's name honors computing pioneer Grace Hopper and that its CPU-GPU integration is the key innovation enabling simultaneous fast and slow Earth system modeling. She also mentions IBM's new quantum computing chip, which integrates error correction on the same chip, and speculates it could eventually accelerate climate modeling further. The hosts discuss whether this resolution improvement will sharpen global warming projections and note that local weather services pull data from centralized government sources rather than running their own models.

As heard by us

A fair, skeptical pass at creatine and cognition, strongest when it separates a possible edge case from brain-health hype.

Jet-lagged but game, Stephen Novella turns this Skeptics' Guide episode toward a useful lesson in how plausible science can shrink under scrutiny.

Read the full review in PlayNext →

Why you'd press play

Want a skeptical take on creatine for brain health?

Read the full recommendation in PlayNext →

Listen to the show on

Apple Podcasts Spotify