i could write a lot about astronomy, particularly this week. jwst time allocation proposals for “cycle 4” (meaning, observations planned for 2025.5-2026.5) are due this wednesday (10/16), meaning everyone at work is feeling the heat, myself included. i’ve written previously about telescope time proposals (here, here) but the pressure—especially for me, staring down the barrel of applying for post-doc positions next year and hoping to bolster my bona fides to appear as enticing as possible to employers and award committees— only seems to be mounting, year after year, as grant and telescope time goalposts continue to shift, the number of applicants grow, and budgets and tenure-track jobs stay flat. i’m sure once the week comes and passes, i can finally relax, right?
there’s plenty to be said, about all the narrowing hallways that extend from my feet, tessellated out in infinite directions. plenty to be said about being an academic hopeful in field vaguely ascribing to the most ancient and profound of human arts. i can’t find the energy to describe any of it with the vigor and preciseness that it necessitates, though. i can only gesture vaguely towards charts put out by the AAS.
the age of slop
one thing i do feel like i have the energy to talk about, though, is slop. webster’s dictionary defines slop as max read writes in new york mag, slop is
“a term of art, akin to spam, for low-rent, scammy garbage generated by artificial intelligence and increasingly prevalent across the internet.”
slop is the “a.i.” (read, probabilistically) generated kindle books overtaking small enclaves of self-published .epub markets on the apple store. slop is shrimp jesus on facebook. slop is the chatgpt’d recipes or explainers or how-tos that begin with their input, “write me a 1000 word blog post about making zucchini pasta,” because the poster didn’t bother to scrub the serial number off. slop is algorithmically generated to satisfy algorithms, an inverted ouroboros, the snake belly up and struggling to swallow. funnily enough, slop has infiltrated the proposal writing rat-race that i find myself stressing out about. i’ve read proposals during “distributed peer review” that were obviously written by chatgpt. there are obvious pressures at play in this area, no one has time to write enough proposals to secure their research, and what about those writing in their second or nth language?
yet, i can’t help but feeling that i’ve been encountering slop for a lot longer than i’ve been forced to deal with so-called artificial intelligence. doesn’t a mr. beast video feel like slop? every second is algorithmic blood being squeezed from a stone. i feel like it’d be more accurate to say that slop is content created or generated for the express purpose of fulfilling an algorithmic niche.
like a good article writer, i’d be amiss if i didn’t refuse to elaborate or back up that claim up, and instead deliberately undermined it in the next sentence to hedge my bets. it’s a dreadful defintion, because it’s way too vague to be useful. by my definition, everything on the internet could be feasibly describe as slop, from a certain point of view.
if i wanted to hit peak article writer, i should have tried to hedge-negate my own claim in the same sentence, but i’m not that pro. yet.
anyways, maybe, slop is content is slop. i imagine it as the krabby patty, which, when bitten into, reveals a gray ooze interior, without origin or substance. i can’t help but imagine that the primary reason that chatgpt has such an easy time recreating the article writer’s affect, such an easy time regenerating explainers or how-tos or recipes, is because the way the internet has been shaped by profit- and attention-seeking algorithms meant that the initial inputs to the large language models were already slop. in the sense that so much of low effort (or undisciplined) writing couches itself in vague, generalizing, citationless claims to pad length and invite the imagination of the reader to fill in the ample blankspace, the medium of llms perfectly matches the content. in the same way that there’s no underlying, deterministic engine (an argument, a point, a perspective, an attempt at style, an exercise in craft or the breaking of rules) driving a shitty book report or explainer article, there’s no deterministic process behind a chatgpt generated block of text, because the text is generated in a series of cascading probabilistic plinko ball horse bounces. ask an “a.i.” chatbot how many times the letter “n” appears in “mewtwo.” text, or content more generally, whose only reason for existing is to meet a deadline or fill an algorithmic niche with more ad-hosting webpages is the gray ooze designed to fill the synthetic krabby patty shell.
to hedge my bets, this isn’t to say i believe “real art” has a “human soul” or something weird and reactionary. i don’t think the proliferation of slop necessitates a bolstering of copyright law (which will only harm free information flow and archival tools, and not actually help the petit-bourgeois artists who are now perpetually flipping out on bluesky over the next edition of dall-e). like any computationally intensive development in technology, it requires a lot of energy resources, which are probably being wasted (under a planned economy, this wouldn’t be a problem, as those resources could be directed towards the positive, constructive uses of the technology and away from the profit-seeking, negative, wasteful uses). i’m sure you can make some funky art or do some useful things with these tools, and they can certainly be used effectively in research contexts. that being said, the hype around these tools is simply a marketing ploy in the vacuum left by the collapse of the crypto industry and the migration of thousands of tech hangers-on towards the ml/ai industry. the world’s engagement with the tools-as-product so far seem to be generally uninteresting, unsurprisingly, as max read’s article details.
what does slop have to do with astronomy, though?
good question, hypothetical reader, thank you for the chance to pivot.
like i said, slop seems familiar. i’ve been wading through the increasingly sloppy internet (sorry) for most of my adult life, llms or no. one of the places that i always seem to encounter low effort content is in astronomy communication on the internet.
astronomy is such an exciting, awe inspiring field of research, and there’s lots of thought, study, and effort in our field that has gone into communicating that research to the world, especially in the context of exoplanets. there’s plenty of ways that can be improved, especially when it comes to public expectation management. see my articles for the planetary society concerning the capabilities of jwst to study the atmospheres of terrestrial and sub-neptunian exoplanets, for instance. there’s a lot there to be improved, and content concessions i had to make during the editorial process, but i’m also proud of how those articles tried to mitigate, and in a few places explicitly push back against, the undue hype about jwst ever “discovering life” in another solar system.
my first brush with sloppy astronomy content was with a classic of the genre, the “i f*cking love science” page on facebook back in the early 2010s. the market hasn’t grown, per-se, but it’s evolved in the context of automative tools. sites like this, for example phys.org, are human supplied/supervised content farms. generally they have a few editors (or now, some automated processes) who scrape press release sites for major universities, research institutions, or space agencies and reprint (or consolidate) the press releases of a specific subfield of research, referring to and linking back to the press release page. it means that when you look up a more obscure result or topic, usually what will come up are a bunch of semi-identical articles all referencing a small press release from a university press office.
for example, i was second author on a paper published in the Astrophysical Journal Letters recently. our team, led by Kyle Franson out of UT Austin and the McDonald Observatory, used jwst to take a picture of a young exoplanet, named ‘af leporis b.’ we wanted to learn more aobut the planet’s atmosphere and measure how bright it was, to learn more about the energy budgets of young planets, which cool after they form, as they radiate away their energy (but how, and by how much in a given time? maybe that can tell us more about their formation and fate!). We managed to observe the planet in a long wavelength filter, but not in two shorter wavelength filters, because the planet is relatively cold and colder things emit more of their light at longer wavelengths.
since it was an impressive observation, techologically speaking—like spotting a firefly next to a lighthouse from across the country—the McDonald Observatory press office wrote up a press release. the press release was picked up by someone at phys.org, and reposted verbatim, which was then reposted to the phys.org MSN feed where it has a solid 22 likes. somehow, the nice press-release image that kyle mocked up (above), didn’t make it through this process of re-posting. the phys.org editor went with a figure from the appendix of our paper, showing a gallery of our non-detections, and a marginal re-detection of the planet with a ground based telescope. Despite Kyle and I’s best efforts, these images look, pardon the french, like shit. I can only imagine the uninformed reader of the phys.org MSN feed squinting at this gallery, this testament to our hubris as observational astronomers, pushing the technology too far and being slapped down by the universe, and skimming the headline that describes this exciting image before liking it and moving on. In their haste to repost this press release to their SEO-hyper-optimized website, an editor confused this image in our appendix for our key result, and now 22 boomers somewhere out in the world have the wrong impression of Kyle’s excellent paper. its not slop, by the strict max read definition, but it feels sloppy.
have you interacted with kid’s youtube slop recently?
(i didn’t wait for you to pose this transition as a question, really sorry about that).
a few years ago, i was researching some protoplanetary disks, and ADS, the service i use to look up papers, was down. i googled a somewhat infamous protoplanetary/”transition” disk system, hd 100546 (that’s the “henry draper” star catalogue, star #100546) and stumbled on a youtube video that drove me to tears laughing. it was posted by a children’s youtube channel, kids learning tube (apparently also an app hosting the same videos). the channel seems to produce animated informational videos on science topics for kids by sort of sing-songingly reading wikipedia articles. i’ve got no problem with wikipedia (that’s a lie, but for astronomy in particular, there are usually enough dedicated citizen scientists trawling and culling the various articles that most of the information there is more accurate than i’d expect). although most of the channel’s videos published recently seem to be slop (by my definition), i think KLT began like most of these things do, with acreator attempting to earnestly create something on the internet. some of the oldest videos on the channel, like this “what is stellar evolution” video, have some real heart to them, even if the animation style is unnerving to me, personally.
now, however, after what i can only assume was a series of hostile takeovers by successively larger production companies, the channel begins its videos for kids by shilling merch that would put redbubble to shame. real “graphic design is my passion” stuff. this ad before the video, which i can only assume was proceeded by a youtube ad, takes up a full 1/6th of the video runtime. the song begins “im a star surrounded by a circumstellar disk, my name is hd one oh oh, five fourty six,” and doesn’t let up. the majority of the song focuses on the “planet” that orbits this star wtihin the circumstellar disk, but there’s room at the end for a shoutout of the southern constellation musca that the star is a member of, and the fact that hst looked at the system a while ago.
the video astounded and horrified me, but to the degree that a normal layperson with passing interest in astronomy or their ipad-affixed child with a hyperfixation about astronomy, there’s nothing wrong with it. it vaguely describes a star, it’s circumstellar disk, it’s planet (hd 100546 b), all while quoting attributes about these things with no real context for why those attributes are important to quote. why do i, the viewer, care that the planet orbits at 55 au from the star? why does it matter that the star is located in the musca constellation? i could explain these things, but the video, cribbing from wikipedia which itself is just a repository for these isolated, wandering facts, cannot. even KLT cannot fight the youtube algorithm’s invisible hand; they now produce a mixture of these short-form videos about weirdly specific phenomena (“Size Comparison Of The 10 Biggest Moons In The Solar System! | KLT” comes in under 3 minutes), as well as the hot, algorithmic pushing multi-hour videos (“Size Comparison Of The 10 Biggest Moons In The Solar System | Space Songs For Kids | KLT” which lasts about 1h10m). like phys.org, this channel itself isn’t bad, or even purposely malicious. KLT isn’t the strongest evidence i could use in my epic takedown of niche astronomy infotainment, in an absolute sense. it’s just another of the billions of attention demanding shortform animations packaged for children’s endless consumption. it’s not “a.i.” generated, and some of it in a vacuum probably wouldn’t even count as slop by my definition. which, kinda does make it the best evidence i could use in my epic takedown, because…
hd 100546 b does not exist.
i have some personal history with this star. i was working on images of the system at the time i stumbled across the KLT video. this work was published in this paper. my undergraduate advsior has an even deeper history with the star, because she had published a paper in 2017 refuting the existence of the planet, hd 100546 b, that the video was describing.
hd 100546 b is a classic case of direct imaging post processing woe: a cautionary tale for the students who came after, an impolite dinner conversation topic for those who lived through it. the host star, like the KLT video describes, is a “star surrounded by a circumstellar disk,” meaning a young star, newly formed (a few million years ago), still wearing a frayed cloak of dust and gas from which it will spin new planetessimals and eventually new planets. these are exciting targets for direct imaging, because if we want to image a planet, it is easiest to image it young, when it still retains all the heat from its formation. the only problem is that all that dust and gas in the circumstellar disk, from which planets are supposedly forming, can be chopped and scrambled in the imaging process and appear like points of light, mimicking how a true planet might appear. this appears to be the case with the “planets” that were discovered around hd 100546. the first warning sign has actually become the weird “quotable” that recurs in the slop that has continued to feature hd 100546 b. it’s too big.
the initial publication, by Quanz et al. (2013), claimed a candidate detection of a point source in the disk in the infrared. it’s a bit sketchy, but worthy of follow up, especially since they only observed the candidate planet in one color of light. there was a flurry of follow-up observations, foremost among them repeated imaging “recoveries” by both Quanz et al. (2015), picture below, who titled their paper “CONFIRMATION AND CHARACTERIZATION…” and by, for instance, Currie et al. (2014) who looked at the system with a different telescope but similar colors (and, funnily enough, show that the circumstellar disk extends right up towards the point source and is rather bright even at these infrared wavelengths). the brightness of this source, compared to its apparent temperature, would require it to have an emitting area many times the size of Jupiter. The problem is that planets don’t get much bigger than Jupiter: even if they’re many time more massive, they just become more dense while maintaining about the same radius. Quanz et al. find a blackbody temperature of ~900K and an emitting of 6.9 Jupiter radii, which you can’t recreate with just a planet. you have to invoke some dust, in this case they claim that this is dust surrounding the planet in a circumplanetary disk that is about 7 Jupiter radii large, but in all honesty, they didn’t have the data to claim that the blob wasn’t just all dust, all the way down.
following studies, like Currie et al. (2016) would claim even more planets existed in the system (consistent with more clumps of dust from the circumstellar disk), until two papers in 2017, Rameau et al. and Follette et al. took the candidate planets down a peg by revealing that they could be explained purely by dust and image processing tricks.
the blob doesn’t even move in orbit around the star like a planet enshrouded in a circumplanetary disk would. the damage (little that it is) had already been done, though, because the planet had been entered into the various archives of exoplanets that exist on the internet.
there’s nothing astronomy slop loves more than an extreme qualifier, and when it was entered into the exoplanet archives as having a 6.9Rj radius, hd 100546 b ensured its longevity.
our first example, from Kyplanet on youtube, isn’t slop. it’s yet another cautionary tale. i’d say Kyplanet is very well meaning, while i disagree with many of their beliefs (many videos about space colonization, many such cases with exoplanet enthusiasts), i really appreciate their vendetta against J1407b, a deep-seated dislike of another dumb exoplanet “candidate” that i also share. that video has a nice little recap of some various slop channels. their other video about how there are “no known habitable exoplanets” is also good. in this way, they appear to significantly contribute to “anti-slop” in the exoplanet space. nevertheless, they are not immune to the insidious case of hd 100546 b. in their video titled “HD 100546 b and the Tartarian Planets,” the content of which i, again, deeply appreciate because it uses hd 100546 b as a springboard to discuss directly imaged planets (yay!). They go so far as to describe the radius problem with hd 100546 b, but the problem here is accepting the existence of the planet at face value. they explain that the “planet” could be as small as neptune with a big shroud of dust surrounding it, which, i think is probably the only case left to be made for hd 100546 b as a real existing astrophysical source, except that i think the much more realistic explanation is the one that’s been repeatedly found in other instances, that is, “clump of dust that was scrambled and pawned off as a planet by excitable astronomers.”
this next one, “Largest Planet In The Universe| HD 100546 B|Universe Science,” is a classic of the astronomy slop genre. it features a bunch of artist renditions smashed together with some adobe premiere zoom and pans, all while some awful stock music that mimics the stranger things theme song blasts your ears. just like the KLT video, it features “facts” that are disconnected from any context or meaning, floating signifiers that the viewer with a passing interest in or current hyperfixation on astronomy can roughly translate in their brain as somehow “important” or “unique.”
our next example is from MetaBallStudios, who specialize in the bread and butter of tiktok-style shortform slop: size comparison videos. why are you laughing.
MBS’ video (no, not him) on hd 100546 features a zoom out, from earth, to Jupiter, to hd 100546 (6.9 times Jupiter’s size, remember?), to the Sun (which is about 10 times Jupiter’s radius). somehow, it seems to cut off too quickly, but at least the music sounds like a hip work-training video.
and yet, even the titans were thrown down.
i managed to find a video claiming a planet even larger than hd 100546. this one fits both my and read’s definitions of slop: “a.i.” generated voice-over, plenty of stock footage stapled together, and complete misinformation. it claims that the planet “KIC 10735564 B” (here, it correctly uses the capital B, which designates an object as the secondary component of a binary star pair) is larger than the sun (and therefore hd 100546 b).
the kepler space telescope was a mission to observe many thousands of stars to look for transits of exoplanets, the light from the star dimming periodically as the planet passes between the star and our line of sight by chance. the “kepler input catalogue” or KIC was a list of all the stars that the telescope would be monitoring. usually, we never refer to any star by it’s name in this catalogue, once astronomers detected a few recurring transits, they’d call a system a kepler object of interest (KOI) and assign it another number. KIC 10735564 corresponds to KOI-3617, the 3,617th kepler object of interest. it didn’t stay interesting for long, because one of the major contaminants in transit surveys are binary stars, which on grazing orbits can mimic the transits of massive planets; KOI 3617 is one such case. the video describes it as “coming in at a whopping 1.44 million kilometers,” in what (radius? diameter? circumscribed distance?) we might never know. at least they’re clear, the planet “does not contain any solid surface.”
so, who bears responsibility for slop? is it the tools that make slop-creation increasingly more accessible? is it the ad-sense fueled algorithms that make a small subsection of slop profitable, thereby inviting thousands of slop creators to ply their trade? is it the science—mistakes, either as a byproduct of the scientific process or the malign work of a few bad actors, spun into press releases by attention hungry astronomers, facilitated by the restrictive funding and advancement structure of the field? it all seems rather low stakes, anyway, with everything else going on in the world. but the slop will continue to flow, dribbling around your feet, burbling past your ankles, gurgling as it envelops your knees.
maybe, one day, you’ll be calling me a slop-hawker. maybe you’ll be right.
Leave a Reply