There are now many reports of a new black hole image generated from the hidden center of the milky way.

My explanation for the new and old "black hole" images:

How it works: The many radio telescope groups joined together to form one huge artificial lens spread over earth. It is a great way to increase the resolution. You take the raw data from the microwave telescopes, and by adding all the waves together you simulate a huge lens.

The noise:
This raw wave data is of course hindered and distorted by weather and atmospheric conditions. And hindered / distorted due to the interstellar matter around the milky-way center.

Sadly there is also systematic error in the way the radio telescopes are positioned. So I am looking forward for the raw data analysis, and the noise reduction algorithms that they used. And how they worked out the real world resolution, without the over-optimistic guessing (that seems to be standard in astronomy to promote new findings)

The bokeh effect as standard error:
The algorithms and expected lack of focus increase the Bokeh effect. This is bokeh: https://i.imgur.com/exgBT9X.jpg

It projects light spots towards the image in the pattern that is placed in front of the lens. It is well-known in photography, but in astronomy we see more likely the cross shaped distortions caused by diffraction.

And this is one red laser dot out of focus: https://i.imgur.com/4OSEHn5.jpg

The 4 points are arranged in the positions of the radio telescope groups, when seen from far away from earth.

This is with a bit better focus. https://i.imgur.com/Nfhz9vV.jpg

And it looks already very similar to the "black hole" So the systematic error in the data has a bias towards the "black hole" shape.

Machine learning to "enhance" the image When machine learning is used, we can be pretty certain that the algorithm has made up non-existent data to work towards the wanted result.
That is how machine learning works. That is why we have sites like "ThisPersonDoesNotExist"
It is as imaginary as CSI enhance.

In both images they needed machine learning to "correct" for all the noise.
They did not just select the data that was without noise and in focus, because that information was unavailable.
You need a good reference to know what is in focus and what is noise at these extreme resolutions.
You can only find out what the noise is compared to the nearest telescopes. And in the first image we even know that they did used data with almost 100% noise compared to the nearest telescopes. So this data was completely useless most of the time. Especially the data that was necessary to focus.

Sky Scholar (Radio imaging expert) on the black hole image.
Black hole image = not science
Scientific analysis
The data fabrication

And here are the options that the algorithm generated from the previous "black hole":
https://i.imgur.com/KPqvIK6.png
These variations were from 4 different groups, and they depend on which part of the noisy data you would prioritize.
From the distance the "black hole" is a very bright object with a beam coming out of it on two opposite sides.
And this is still an option, but was probably not in the machine learning database.

Here is a video: VFX artists debunk CSI enhance effects.
https://www.youtube.com/watch?v=jT2sAz3e2yc

This all means that we need a damned good skeptical reality check of the raw data and the algorithms.