Cool discovery, but I'd be real curious about the false-positive rate when letting coding agents loose on raw astronomical telemetry. Hallucinations in data reduction are no joke.
> TESS comes back to this part of the sky in November. I submitted an observing proposal asking it to record this star every 2 minutes while it's there.
If you can set the criteria for testing that a human would also use, then any llm can do the job.
What is uncool is how people are using cloud based AI and their universities are not providing self hosted models that keep data within the university. Cloud AI is useless due to it being no different than publishing everything you do publicly.
Plot twist: The discovery is valid. But we figure out the author was not a innocent PM, but a team of astronomers working for Anthropic, And Anthropic's marketing department come up with the idea of crediting it to a PM. :-D
Dismissing something for lacking formal academic review is counter to the ethos of the site, and the link itself mentions that verification is coming in a month and there's at least a loose consensus of actual people that this might be valid.
If they'd said "excited to see if this replicates in November!" I would have at least believed that they'd read the link and/or had some interest in the topic.
I don't see any value in "this topic doesn't interest me", though
It's interesting that a few months ago your sentiment would have been upvoted. I've noticed recently that things seem to be flipping around. Sadly, I think it's inevitable that people will accept a form of communication that consists of people trading slop. Even worse, the author lies about using AI for the writeup in the comments.
Commenting about AI tone in an article is the new commenting about the contrast between the font color and the background being too narrow. Nobody wants to hear it over and over and over again on every thread.
On the other hand, I get mentally exhausted reading AI slop over and over again.
My brain can't help itself in looking for AI tells and gauging if writing sounds like AI or not.
Then once it latches onto the idea that the writing is probably AI, then it feels like half my brain is just reading the text looking for the AI smells.
Calling it "AI smells" like "code smells" might actually be what is going on in my brain, and I start code reviewing it at that level rather than reading it for what it does.
I really would prefer people not just post slop. I don't care if you use it to clean up your own writing, but don't just let it rip.
You're right that nobody wants to hear it, but I think that's because nobody wants to find out over and over and over that every thread is AI generated. Not because when something is AI generated, they wouldn't like a heads up so they can find something else.
For example, I've had people thanking me for pointing out LLMish later in an article when it's not as obvious at the beginning. It saves the time of thinking the article is manually written when it'll later turn out to be AI instead.
Note that I don't necessarily have anything against AI in particular, but rather the chronic lack of information density typical of LLM-generated articles. Some articles that were LLM-assisted are still information-dense enough to keep me satisfied.
As someone who hates reading AI-written stuff (stylistically, I'm sure that someone could build an LLM that writes well) I have begrudingly accepted that it's here to stay.
However, I think that AI-slop is genuinely harder to read and less information-dense than stuff written by a person, and I catch myself abandoning articles and other written content early on when they smell too much like AI. It's simply less worth it to read them because I'll have to spend some precious time figuring out wtf the text is actually trying to say.
It seems like the flow is "User A writes bullet points and has a model convert them to prose to send to User B. User B then uses a model to convert the text back to bullet points."
I would in every case prefer the bullet points or raw transcript of the AI session to just the final output.
It is harder to read, and it's especially egregious in the human-facing documentation or code comments.
It takes considerable effort, rules, examples, hooks and prompt to keep the popular $model reasonably concise, and yet STILL from time to time it starts vomiting absolute garbage.
At the moment the best recalibration is "shaming" it and showing both the negative (the vomit) and positive (slop replaced with human sentences). Good for a moment (up to 200k context).
My good sir, I regret to inform you that the informality of the English language has been in decline for quite some time. Possibly since that vulgar bard Shakespeare abandoned all pretense of a shared language and raised a generation that thinks words like "fashionable"[sic!!] are real words.
However, it is notable that the rules of this particular venue request one be charitable to your fellow forum goers, and thus the admonition that their preferred communication is "slop" is generally considered unwelcome.
Would that you had only been able to word your own communication is in proper vocabulary and grammar; alas, I must conclude that even as you reject modernity, you have embraced the "slop" of the past.
But to address your last sentence: OP explicitly admits to using AI in the Reddit comments, and apologizes for it. https://www.reddit.com/r/ClaudeAI/comments/1wzw8zd/i_think_i.... If you are going to repeat this libel, at least have the decency to provide some sort of citation
is this supposed to be an example of bad writing because unnecessary bombast is definitely a sign of it
good writing is not just replacing random words with entries from thesaurus.com, it's writing for your reader. that's one thing LLM-generated text is very bad at - frequently violates the show-don't-tell rule, it lacks any interiority by it's very nature, etc
if you don't practice writing yourself, you don't develop the muscle of knowing what your reader's expectations are and you will forever remain a mediocre writer
I assume myself and others like me since it is being posted on a forum where each and every comment has not only an audience but a points-based rating system
Regardless of authorial intent, everything posted here is for the wider HN audience
I will concede I had assumed a more educated audience, which could pick out reference to a particular linguistic style, rather than confusing it for a wild joy ride through thesaurus.com
Conversely, if you yourself write solely for points, then you might as well be a machine - for all reasonable virtues of truth, community, and free expression will find themselves at odds with such petty incentives. I am happy to lose a karma or two to speak my own mind, rather than playing puppet for your upvote.
It reads as an example of English writing that was once considered "proper". The point I see is that it shows how this complaint about AI writing is history rhyming with itself again.
There is a difference between objections to style and objections to clarity and conciseness. My complaint with AI isn't that it's not formal enough, it's that there's so much bullshit and so little content in AI-generated blocks of text.
"Slop" is generally meant as a derogatory label against anything in the AI vernacular - it is not an objection to clarity, it is a repudiation of the idea of using AI for writing at all.
It's the same tired game we play, as a society, any time a new writing style enters the mainstream. That game goes back to at least Shakespeare's era, where our illustrious bard was called out for his tendency to make up words ("fashionable" being, I believe, an actual example of this?)
Not an astronomer, but I'm a non-programmer who spent two months building a fairly big web app with Claude, so the false-positive question is the one I'd ask too.
What saved me wasn't asking Claude whether it was right (it almost always says yes). It was checks that don't depend on its own report: run it on data where you already know the answer, and give a fresh session only the result and ask it to break it.
For a planet hunt I guess that means hiding a few fake planets in real data to see if it finds them, and running the same pipeline on shuffled data to see how many "planets" it finds in pure noise.
I've been tinkering with the latter for the last couple of months. Yesterday my system made some progress on instabilities on a particular superconductive system: https://colinocallaghan.com/autonomous-ai-research/Spatial-P...
> "We got vibe astronomy before GTA6"
https://getartcraft.com/apps
Friend sent me this one, kinda glad people are learning what I learned with Claude 4.6… you can build anything at this point.
> even got NASA to point a telescope at it for confirmation
> TESS comes back to this part of the sky in November. I submitted an observing proposal asking it to record this star every 2 minutes while it's there.
> It was approved. Program #100.
What is uncool is how people are using cloud based AI and their universities are not providing self hosted models that keep data within the university. Cloud AI is useless due to it being no different than publishing everything you do publicly.
"Vibe-turfing"?
Asking for a friend.
Dismissing something for lacking formal academic review is counter to the ethos of the site, and the link itself mentions that verification is coming in a month and there's at least a loose consensus of actual people that this might be valid.
If they'd said "excited to see if this replicates in November!" I would have at least believed that they'd read the link and/or had some interest in the topic.
I don't see any value in "this topic doesn't interest me", though
I DESPISE this kind of slop prose
My brain can't help itself in looking for AI tells and gauging if writing sounds like AI or not.
Then once it latches onto the idea that the writing is probably AI, then it feels like half my brain is just reading the text looking for the AI smells.
Calling it "AI smells" like "code smells" might actually be what is going on in my brain, and I start code reviewing it at that level rather than reading it for what it does.
I really would prefer people not just post slop. I don't care if you use it to clean up your own writing, but don't just let it rip.
It's not AI tone, that's not a thing, it's just low quality text.
For example, I've had people thanking me for pointing out LLMish later in an article when it's not as obvious at the beginning. It saves the time of thinking the article is manually written when it'll later turn out to be AI instead.
Note that I don't necessarily have anything against AI in particular, but rather the chronic lack of information density typical of LLM-generated articles. Some articles that were LLM-assisted are still information-dense enough to keep me satisfied.
However, I think that AI-slop is genuinely harder to read and less information-dense than stuff written by a person, and I catch myself abandoning articles and other written content early on when they smell too much like AI. It's simply less worth it to read them because I'll have to spend some precious time figuring out wtf the text is actually trying to say.
It seems like the flow is "User A writes bullet points and has a model convert them to prose to send to User B. User B then uses a model to convert the text back to bullet points."
I would in every case prefer the bullet points or raw transcript of the AI session to just the final output.
It takes considerable effort, rules, examples, hooks and prompt to keep the popular $model reasonably concise, and yet STILL from time to time it starts vomiting absolute garbage.
At the moment the best recalibration is "shaming" it and showing both the negative (the vomit) and positive (slop replaced with human sentences). Good for a moment (up to 200k context).
However, it is notable that the rules of this particular venue request one be charitable to your fellow forum goers, and thus the admonition that their preferred communication is "slop" is generally considered unwelcome.
Would that you had only been able to word your own communication is in proper vocabulary and grammar; alas, I must conclude that even as you reject modernity, you have embraced the "slop" of the past.
But to address your last sentence: OP explicitly admits to using AI in the Reddit comments, and apologizes for it. https://www.reddit.com/r/ClaudeAI/comments/1wzw8zd/i_think_i.... If you are going to repeat this libel, at least have the decency to provide some sort of citation
good writing is not just replacing random words with entries from thesaurus.com, it's writing for your reader. that's one thing LLM-generated text is very bad at - frequently violates the show-don't-tell rule, it lacks any interiority by it's very nature, etc
if you don't practice writing yourself, you don't develop the muscle of knowing what your reader's expectations are and you will forever remain a mediocre writer
Regardless of authorial intent, everything posted here is for the wider HN audience
Conversely, if you yourself write solely for points, then you might as well be a machine - for all reasonable virtues of truth, community, and free expression will find themselves at odds with such petty incentives. I am happy to lose a karma or two to speak my own mind, rather than playing puppet for your upvote.
It's the same tired game we play, as a society, any time a new writing style enters the mainstream. That game goes back to at least Shakespeare's era, where our illustrious bard was called out for his tendency to make up words ("fashionable" being, I believe, an actual example of this?)
What saved me wasn't asking Claude whether it was right (it almost always says yes). It was checks that don't depend on its own report: run it on data where you already know the answer, and give a fresh session only the result and ask it to break it.
For a planet hunt I guess that means hiding a few fake planets in real data to see if it finds them, and running the same pipeline on shuffled data to see how many "planets" it finds in pure noise.