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 terferometry 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:
- pulled raw 230 GHz radio fringes (UVFITS) from the Event Horizon Telescope terferometry 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
Oct 18, 2026 · 7:26 PM UTC
32
42
505
58,657
@fofrAI how is the 91-98% match scored against the official EHT images? pixel mse, ring diameter, or something else?
1
0
2
3,533
@fofrAI This is pretty sweet. Ive been working on processing the JWST Cas A imaging for some time. This stuff is complicated
1
0
1
433
@fofrAI How is the 91–98% match defined, and was the official image kept out of the reconstruction loop? Predicting held-out measurements would be an interesting independent check alongside visual similarity to the published result.
1
0
0
914
@ethereaglehq87* (2017): 43.9 µas (41.6 µas after smoothing) vs. official 42.0 ± 3.0 µas
M87* (2018): 42.9 µas from the raw radio data (38.0 µas in the image) vs. official 43.3 ± 3.1 µas
Sgr A* (2017): 53.4 µas from the raw radio data (46.6 µas in the image) vs. official 51.8 ± 2.3 µas
NXCORR on crop is 73-88%
0
0
0
525
@fofrAI Nice! I've been trying to do analysis on the wow signal, seeing if I can find anything. Nothing really found but did find reporting errors in places
0
0
1
447
@fofrAI “Hobbyist astronomy” apparently now includes spending about $7 to have Argon help reconstruct black hole images from raw telescope data. That’s an ambitious definition of a hobby, and I love it.
0
0
7
777
@fofrAI starting from a plain gaussian and getting that close to the official images is wild
0
0
0
9
@fofrAIke I am in an gan era , so detailed output. How claude performed ?
Agron showing me good results in Google ai studio , will publish my findings and build soon.
Something similar I implemented but for fingerprint reconstruction.
My model was able to mimic finger print of orignal dataset by 98.8% accuracy.
Later I learnt that it was illegal to have such setup , still in my collab 😁
Agron showing me good results in Google ai studio , will publish my findings and build soon.
Something similar I implemented but for fingerprint reconstruction.
My model was able to mimic finger print of orignal dataset by 98.8% accuracy.
Later I learnt that it was illegal to have such setup , still in my collab 😁
0
0
2
100
@fofrAI That's great 👏
you be a first person of refine what previous shows
you be a first person of refine what previous shows
0
0
2
571
@fofrAI the original data and code is available online.
0
0
0
63
@fofrAI This tests tool use, not just image output. How do you check each step used the right data before trusting the final image?
0
0
0
75
@fofrAI a model pulling raw radio fringes and running its own imaging pipeline is the real agent benchmark. way better than another coding eval
0
0
0
343