Astronomy · stargazing · space science

The sky is out there. Start gazing.

Celestial event previews, plain-language science explainers, and honest gear notes — written by an amateur astronomer in Cyprus. 3–5 articles a week. No cosmic ballet.

The faint giant low-surface-brightness galaxy Malin 1, a sprawling pale spiral against a field of smaller galaxies, imaged in colour by the DESI Legacy Imaging Survey, one of the sky surveys used to train astronomical foundation models.

AI × astronomy

Foundation models came to astronomy — and one matched a supervised network with 1% of the labels

Here’s the number that made me put my coffee down. A model trained on Euclid galaxy images matched a fully supervised network at estimating galaxy redshifts and stellar masses — using 1% of the labeled examples. Ninety-nine percent of the labels thrown away, same accuracy. That result is from the Euclid Quick Data Release paper by Siudek et al., published in Astronomy & Astrophysics on 30 June 2026. It’s one of three papers from the last two years that convinced me astronomy has quietly acquired what the machine-learning world calls foundation models: a single large model pretrained once on a mountain of unlabeled data, then adapted cheaply to many different tasks. ...

The faint interstellar comet 3I/ATLAS as a small diffuse smudge among background stars, imaged the day after its discovery.

Space science

3I/ATLAS: Webb's spectra point to an interstellar comet older than the Sun

I’ve now watched three objects from other star systems pass through ours, and only one of them said anything worth hearing. The first, 1I/ʻOumuamua in 2017, was gone almost before we understood what it was: a fast, tumbling sliver that never grew a tail and left us arguing about its shape. The second, 2I/Borisov in 2019, was a proper comet with a coma and a tail, alien in origin, but the kind of object we already know how to study. The third is 3I/ATLAS, and it did what the other two wouldn’t. It stayed in front of the James Webb Space Telescope long enough for Webb to read its chemistry line by line, and the readout points to a comet that froze into existence before the Sun did. ...

NASA's Pegasus barge, carrying the Roman Space Telescope in a white shipping container, moored at the Launch Complex 39 turn basin with the Vehicle Assembly Building behind it

Space news

The Roman Space Telescope launches August 30 — a survey machine, not a telescope you point

NASA’s Nancy Grace Roman Space Telescope reached Kennedy Space Center on June 21, arriving by barge at the Launch Complex 39 turn basin, the same channel that once floated Saturn V and Shuttle hardware to the pad. It’s now in the Payload Hazardous Servicing Facility being prepped for a Falcon Heavy launch no earlier than August 30, roughly eight months ahead of its original schedule. That last detail is rare enough for a flagship NASA mission to deserve its own headline. ...

Numerical-relativity simulation showing two black holes spiralling together above a warped grid of spacetime, with the emitted gravitational waveform rippling outward

Space science

A decade ago LIGO caught one black hole collision. The catalog now holds 390.

On 14 September 2015, two detectors in Louisiana and Washington State registered a rising chirp that lasted about a fifth of a second: two black holes of roughly 36 and 29 solar masses spiralling together 1.3 billion light-years away, their final orbits stretching and squeezing the ground by less than the width of a proton. That single event, GW150914, opened gravitational-wave astronomy and won a Nobel Prize. On 26 May 2026, the LIGO–Virgo–KAGRA (LVK) collaboration released GWTC-5.0, the fifth edition of its Gravitational-Wave Transient Catalog. The running total of detected events since 2015 now stands at 390. A pile that size changes what the field can do. Gravitational-wave detection has crossed from hunting individual events into doing population statistics: this is a black-hole census now, and the population is starting to talk back. ...

Wide-field photo of the summer Milky Way stretching from Vulpecula down to Scorpius, its bright star clouds crowding toward the galactic center low in the sky

Beginner stargazing

How to find the center of the galaxy with your own eyes this July

On a dark night this month you can stand in a field, face south, and look almost exactly toward the supermassive black hole at the center of our galaxy. No telescope. No binoculars, though they help. Just your eyes, a clear southern horizon, and enough patience to let them adjust. The direction sits low in the sky between two constellations that are up all July: Scorpius and Sagittarius. You won’t see the black hole itself (its light is buried behind thousands of light-years of dust), but you can point your face down the barrel toward it and see the thickest, brightest star clouds of the whole Milky Way piled up in that direction. From my usual dark spot in the Troodos mountains, that stretch of sky is the best thing summer offers. Here’s how to find it. ...

Kirkjufell mountain and the Kirkjufellsfoss waterfall on Iceland's Snæfellsnes Peninsula, one of the regions inside the August 12, 2026 path of totality

Astrotourism

The August 12 eclipse in Iceland: where to stand, and the weather gamble

On August 12, 2026, the Moon’s shadow makes landfall in western Iceland before it reaches anywhere else you can practically stand on. It crosses the Westfjords, the Snæfellsnes Peninsula, the northern edge of Reykjavík, and the Reykjanes Peninsula, then leaves the island and heads for northern Spain. If you want the longest totality this eclipse offers on land, a hair over two minutes, and you want the Sun a comfortable 24° above the horizon instead of skimming the sea, Iceland is the place. The catch is the sky. Iceland in August is one of the harder clear-sky bets on the whole eclipse track, and pretending otherwise would be doing you a disservice. ...

A man and a woman watch the 2017 total solar eclipse through cardboard eclipse glasses in Madras, Oregon

Celestial event

How to watch the August 12 solar eclipse safely (and what you'll actually see)

From my balcony in Nicosia I’ll see nothing on August 12. The Moon’s shadow doesn’t reach the eastern Mediterranean, so from Cyprus the Moon never crosses the solar disc at all. This is the guide I’d hand a friend who can get themselves somewhere the eclipse actually happens, whether that’s the path of totality across northern Spain or a deep partial from a back garden in London. One rule matters more than everything below it: until the Sun is 100% covered, you look through a certified solar filter or you don’t look at all. Not at 90 percent, not at 99. And if you aren’t standing inside the narrow path of totality, that means filters on for the whole eclipse, first bite to last. Everything else here is built around that line. ...

Infrared image of a nebula in the constellation Scutum captured by NASA Spitzer Space Telescope

Gear

Imaging at 35°C: when a cooled astrophotography camera is worth the money

On a June night in Nicosia, ambient 30°C at 11 pm with no breeze, I ran a 40-minute stack on the eastern Veil Nebula with the Seestar S50. The final image looked passable at full zoom-out. At 100% crop the background was peppered with hot pixels, and the faintest filaments had disappeared into a warm, grainy haze that no processing could salvage. The same target in March, with nighttime temps around 14°C, had noticeably more contrast in those faint regions. Same scope. Same integration time. Same processing. The only variable was temperature. ...

A 63×63×20 Mpc slice through the z=0 snapshot of cosmological N-body simulations at three resolutions

AI × astronomy

Transfer learning made cosmology 10× cheaper — then started confusing new physics with old

A paper published in the Journal of Cosmology and Astroparticle Physics on June 10 asks a deceptively simple question: if you train a neural network on the standard cosmological model and then ask it to look for physics beyond that model, does the pretraining help or hurt? The answer, from a Princeton and Flatiron Institute team, is both. Transfer learning — pretraining on cheap ΛCDM simulations, then fine-tuning on expensive beyond-ΛCDM runs — cut the number of simulations needed by more than a factor of 10 in some scenarios. But it also introduced a failure mode the authors call negative transfer: the network learned the standard model’s patterns so thoroughly that it confused genuinely new physics with familiar old parameters. ...

Space news

NASA's Swift telescope is falling out of orbit. A startup had nine months to save it.

The Neil Gehrels Swift Observatory has spent nearly 22 years detecting gamma-ray bursts, X-ray transients, and ultraviolet afterglows from low Earth orbit. It’s one of the most productive space telescopes ever launched — over 1,600 gamma-ray bursts catalogued, thousands of follow-up observations forwarded to ground-based telescopes within seconds. And it’s falling. No later than Tuesday, a refrigerator-sized spacecraft built by a startup in nine months will ride the last Pegasus XL rocket ever launched in an attempt to grab Swift, lock onto its structure with three robotic arms, and push it back up. If it works, it’ll be the first time anyone has docked with a satellite that was never designed for servicing. ...