Daniel Banasik

Software architect, instrument maker and „guy into robotics”. Focus: robotics and AI in education. Hobby: cosmology and theoretical physics.

[Daniel Banasik @DanBanasik]

One geometric law, three scales. The same acceleration floor a* fits 12 SPARC rotation curves (dwarfs→spirals), 12 Chandra cluster profiles at 0.03–0.14 dex, and the galaxy RAR — where late/early types split by a single parameter A=√(1+f). All within linearized GR.

Preprint available: https://t.co/kK701rj6RN

[Daniel Banasik @DanBanasik]

It seams to work pretty well on 2 of my papers and not the third one. Tested on AI generated abstracts so word count was under 500 in all 3 cases and prose correctly marked as AI. The interesting part is that the paper where I describe 2 based on unit system which is one of my oldest ideas tripped just slightly over the line into AI Ideas. Why would that be? Isnit because it is practically a „unit system refactor” and detector see to much structure mapping?

[Daniel Banasik @DanBanasik]

@CoutureTroll @Marco777Polo @NTFabiano arXiv introduced limit on submissions. This is what I meant by „seams to be buying into this bottleneck scenarios”. It is not a limit which will hit hard right now, but also it is not a solution of the problem.

https://t.co/Pkl8kqvyLz

[arXiv.org @arxiv]

arXiv has updated our policy on rate limiting for all submitters.

This update was made to fairly distribute moderator time & support the arXiv community of staff, volunteers, readers & authors.

Please read our announcement to learn more: https://t.co/lvzWuiv0hx

[Daniel Banasik @DanBanasik]

@florianederer No wonder. Economics is the only science where you can miss the prediction by a 10^10 and still be published. 🥸

[Daniel Banasik @DanBanasik]

Yes, obviously with AI one can prepare the paper much faster. But if it wouldn’t be for the competitive environment driving people to publish asap, before the others , we wouldn’t see this happening so much.
We could as easily use AI in accelerating research without this artificial issue of hundreds papers weekly. More time would go into refinement, cooperation and we would have more steady pace with better outcomes.

[Auyon Siddiq @auyonymous]

I am pleased to announce a new professional milestone: Professor Polson was sufficiently unhappy with my post, and the ensuing Washington Post article, that he decided to email *my Dean* an AI-generated review of my body of work.

1/

[Daniel Banasik @DanBanasik]

@auyonymous Let me just grab my popcorn. This is too good.

[Daniel Banasik @DanBanasik]

@dumierhan Open .md file or „profile -> instructions” panel if you are on Web/Mobie and write the:
Before reply always check the time window since last interaction.
Or something similar.

[Daniel Banasik @DanBanasik]

I have rather different experience actually. I did not send a paper to review but did review it myself with a different model. It found errors and inconsistencies. Then I presented this review to the „author-model” and asked to analyze it.
It did exactly what what your expectations are. It agreed and corrected the errors, discussed other points and simply rejected some of them while partially agreeing with some.
This behavior is consistent. If you cannot get it out of LLM then prompt it in a different way.
„Give me reliable, critical review”
„This is review I received - analyze which points are correct”

But if I prompt iit: „These are peer review comments.” The models usually will try to adding them all.

[Daniel Banasik @DanBanasik]

Are we talking about the same word?

„Bigotry is the stubborn, unreasonable attachment to one's own beliefs and an intense intolerance or dislike of people who have different backgrounds.”

If I write a paper where I present a new argument on some subject and I write it with extensive use of AI because I’m writing technical paper in English, then you dismiss my argument because you would prefer for me to spend much more time with an English dictionary rather than constructing the argument itself - well, Sir, you have just exhibited unreasonable attachment to your own belief that you know better what I should have been spending my time with and dislike towards me coming from non-English background.

You see, I can bend it my way to.
Have a nice day.

[Daniel Banasik @DanBanasik]

Nice. You brought up the word in a first place, not me.
I wrote „It is pure genetic fallacy to dismiss somebody’s entire argument based on AI usage.” And I stand by it.
Wanting researchers to go through the writing process is your preference, not bigotry. Judging an idea's truth by looking at where it came from - that’s logical error.

[Daniel Banasik @DanBanasik]

@cheaf25master Absolutely. All it takes is to be curious about „why this way”. You already have LLM in front of you so the question is going to be not only answered but immediately explained on an example you just vibe coded.
Is there a better way?

[Daniel Banasik @DanBanasik]

@danielgerling @yevgets Yes, I do. If we allow ourselves to completely dismiss an argument solely on the fact of AI being used during writing it then for me this is good example of origin-based judgement. And that is bigotry.

[Daniel Banasik @DanBanasik]

Ok. While reading the prose we are typically looking for the main idea, motivation, what the result means, why this approach and how the pieces connect, right?
Then what is wrong with using LLM to flag gaps, tighten text, clarify notation and perhaps even rephrase so it is easier to understand?

[Daniel Banasik @DanBanasik]

Time spent by author is not a measure of quality. One can write quick and well or terribly for painfully long time.
Nothing will replace the fact that current model of unpaid peer review is completely misguided. You are asked to deliver the value for free to and by Publisher who gets paid for your work.
Absolute garbage model.

[Daniel Banasik @DanBanasik]

@matthias_mrc Remove the „VIBE” from the post and it is still absolutely true. What does it tell you?
9 out of 10 startups failed long before AI.

[Daniel Banasik @DanBanasik]

@mosasaurus27 Achievement is real, science progress is real, benefits may be multiple and I still can’t help thinking: would this rat be me in case aliens come? 🤔

[Daniel Banasik @DanBanasik]

When I read scientific paper, the thing I focus on is content. Writing style may help or be an obstacle. This was always true, AI brought nothing new to the subject here. So when today I hear „it feels like AI writing therefore it is slop” and person completely dismiss the content because of their personal opinion on what good writing style supposed to be, then I say they misplace their attention. It is not writing style whichnis the most important here but the scientific content of the paper.

[Daniel Banasik @DanBanasik]

@komal_uk01