Pinned Tweet
Argon has this amazing natural understanding of the world. It can do so many things from first principles.
Here's a pelican riding a bike, using fourier epicycles.
Video made by Argon.
Time-synced Lyria 3.5 music prompted by Argon.
Gemini 4 Argon is live.
blog.google/innovation-and-a…
There's a lot to show you.
Very cool 3D demo with Argon
Some intermediate outputs, an early pipeline on the left, a refined and iterated pipeline on the right.
I asked Argon to use NXCORR comparisons between pipeline outputs and official images to automatically iterate.
Hey Argon, make me some awesome space timelapses, stitch them together in a video with title cards and a soundtrack
I've been doing some hobbyist astronomy with Argon. Here's one of my experiments, where I tasked Argon with reproducing the first black hole images (from 2017/18) using the same raw data.
Argon:
- pulled raw 230 GHz radio fringes (UVFITS) from the Event Horizon Telescope archives.
- built an interferometry solver using 3-dish closure phases to cancel atmospheric noise
- started each run from a plain Gaussian blob with no ring templates or masks
- recovered M87* (2017 & 2018) and Sgr A* at a 91% to 98% match of the official EHT images
Argon identified the candidates and gave this description of what it did for each:
V838 Monocerotis light echo (0:00–0:08 · May 2002 to Sep 2006): Co-registered 10 archival Hubble ACS/WFC 3-band (F435W, F606W, F814W) drizzled FITS epochs onto a common astrometric grid and interpolated strictly between the bracketing observed exposures.
Supernova 1987A shockwave ring (0:08–0:16 · Sep 1994 to Jun 2016): Aligned 18 archival Hubble WFPC2, ACS/HRC, and WFC3/UVIS 3-band FITS epochs onto the two companion stars and flux-normalized across camera upgrades to show the real blast wave lighting up ~30 ring hot spots.
Cassiopeia A supernova remnant (0:16–0:24 · Jan 2000 to May 2019): Reprojected 13 archival Chandra ACIS-S broadband X-ray FITS observations onto a shared tangent-plane WCS grid pinned to the central neutron star, with exposure-normalized interpolation between consecutive ObsIDs.
Crab Nebula supernova shell (0:24–0:32 · Nov 1951 to Feb 2024): Registered 9 real archival FITS epochs (1951 Palomar POSS-I through 2024 Hubble WFC3) onto a Gaia DR3 WCS grid and interpolated strictly between adjacent observed plates after removing the old 80% 2024 back-weighting and synthetic wisp wave.
Tycho's 1572 supernova remnant (0:32–0:40 · Sep 2000 to Aug 2021): Clamped strictly to the 6 real Chandra ACIS X-ray FITS epochs (ObsID 115 in 2000 through ObsIDs 23539/24405 in 2021, removing the pre-2000 extrapolation and analytical cosine formula) and interpolated between consecutive X-ray maps using empirical sector proper motions.
Eta Carinae Homunculus Nebula (0:40–0:48 · Jun 1995 to Feb 2020): Co-registered 9 archival Hubble WFPC2/PC and WFC3/UVIS narrowband/UV (F658N/F656N and F336W/F410M) FITS epochs onto the central binary to track the 25-year ballistic expansion of the bipolar dust lobes.
R Aquarii precessing jet (0:48–0:56 · Oct 2013 to May 2023): Clamped strictly to the 6 real Hubble WFC3/UVIS narrowband ([O III], Hα, [O II]) FITS epochs inside the unclipped detector footprint, with all pre-2013 extrapolation, 180° mirrored fill, and synthetic precession/Mira formulas removed.
Wolf-Rayet 140 dust shells (0:56–1:04 · Jul 2022 to Sep 2023): Clamped strictly to the 2 real James Webb MIRI 3-band (F770W, F1500W, F2100W) FITS epochs (ERS-1349 on 2022-07-27 and GO-3823 on 2023-09-09), removing the 1992–2024 phase-wrapped extrapolation to show the real 0.38″ 13.4-month shell expansion.
Quasar 3C 273 relativistic jet (1:04–1:12 · Jul 1995 to Nov 2024): Restored all 156,094 real VLBA 15.4 GHz AIPS CLEAN components across 140 MOJAVE FITS epochs with a uniform 0.60 mas circular Gaussian beam (w_knot = 0.0, zero synthetic knot overlays).
Binary black hole candidate OJ 287 (1:12–1:20 · Apr 1995 to Dec 2024): Restored all 18,913 real VLBA 15.4 GHz CLEAN components across 195 MOJAVE FITS epochs with a uniform 0.45 mas circular beam and direct MJD-bracketing interpolation (w_knot = 0.0).
Blazar BL Lacertae magnetic jet (1:20–1:28 · Apr 1995 to Dec 2024): Restored all 62,239 real VLBA 15.4 GHz CLEAN components across 245 MOJAVE FITS epochs with a uniform 0.45 mas circular beam (w_knot = 0.0, zero synthetic ridge-line wave multipliers).
Perseus Cluster core 3C 84 (1:28–1:36 · Oct 1995 to May 2026): Restored all 57,538 real VLBA 15.4 GHz CLEAN components across 91 MOJAVE FITS epochs with a uniform 0.40 mas circular beam over an 8.0 mas field of view that keeps the expanding southern C3 lobe unclipped.
M87 parsec-scale radio jet (1:36–1:44 · Jul 1995 to Feb 2026): Restored all 34,922 real VLBA 15.4 GHz CLEAN components across 48 MOJAVE FITS epochs with a uniform 0.60 mas circular beam (w_knot = 0.0, removing the analytical spine_hollow and map_rails multipliers).
Barnard's Star on sky-survey plates (1:44–1:54 · Jul 1950 to Jan 2012): Co-registered 6 unmasked historical sky-survey FITS plates (1950 Palomar POSS-I, 1982 Quick-V, 1988/1991/1993 POSS-II, and 2012 Pan-STARRS1) onto a shared WCS grid, removing the np.nan star mask and synthetic Gaia PSF overlay so the real photographic star image moves 637″ across the plates.
I've been doing some hobbyist astronomy with Argon. Here's one of my experiments, where I tasked Argon with reproducing the first black hole images (from 2017/18) using the same raw data.
Argon:
- pulled raw 230 GHz radio fringes (UVFITS) from the Event Horizon Telescope archives.
- built an interferometry solver using 3-dish closure phases to cancel atmospheric noise
- started each run from a plain Gaussian blob with no ring templates or masks
- recovered M87* (2017 & 2018) and Sgr A* at a 91% to 98% match of the official EHT images
Trying to make art from astronomy data. This shows 545 of our long-period comets, and our three dramatic visitors.
"Three Strangers"
Love this demo
Gemini 4 Argon - Recreating the interactive lens lab demo by @RyanSael !
I gave it the post video and asked it to make me something similar and i think it nailed it...
Made fully in three.js and it's live at thtbee.github.io/waggles/pla…
It took 45 minutes for the whole thing..!
One of the intermediate outputs from this animation process, where I first tried adding some color while also testing how Argon understands dynamic motion.
I found Argon could sketch pretty quickly when using a custom dsl tool it had built for the job.
So I asked it to try animating a scene by sketching the next few frames from scratch each time, rocking the motion back and forth a bit, and iterating to make a short clip.
Then I asked it to add cel shading and transparency and to put it all together onto a background.
An experiment in making music with math and Gemini 4 Argon, very much inspired by AA's excellent work and @MazQmusic as well.
All made using an agent in Antigravity CLI.
