What do boys like? Electolytes?
Cosmic Funnies
Misplaced Lens Cap
RMH
I'd rather be in outer space 🛸
trying on a metaphor
Show & Tell
Keni
Not today Justin

JVL

titsay
Today's Document
noise dept.
Peter Solarz
Stranger Things
Monterey Bay Aquarium
official daine visual archive

Love Begins
2025 on Tumblr: Trends That Defined the Year
$LAYYYTER

if i look back, i am lost
seen from Italy
seen from Germany

seen from United Kingdom
seen from Switzerland
seen from United States

seen from United States
seen from United States
seen from Philippines
seen from United Kingdom

seen from United States
seen from China
seen from TĂĽrkiye

seen from United States

seen from China

seen from Malaysia
seen from United States
seen from France

seen from Malaysia
seen from United Kingdom
seen from United States
@the-arigen
What do boys like? Electolytes?
I love characters who are extremely hypercompetent and simultaneously massive losers
mooooom they're arguing about polish mathematicians in the memes chat again
they shouldnt put video games on the switch
I am once again digging through a set of four primary sources only available in a language which I do not effectively speak and only read with a dictionary and a notebook in order to accurately represent a mythological creature including its cultural connotations and make sure I'm not being disrespectful about it for my little fantasy story very few people are likely to ever read
this writing thing is going to kill me
sometimes i hate on a popular thing but then i forget that most people hate it for standard reason A, and that i neglected to clarify that i hate it for secret bizarrely specific reason P, and that i think it’s mean to hate on it because of standard reason A which is honestly really unfair, but like. secret bizarrely specific reason P is kind of hard to explain
/*completely incomprehensible poly relationship drama cut for brevity*/
... this is gonna be my whole day tomorrow, isn't it?
update on this: no longer tired, tipsy, and ticked off, I realized that only some of this was something I cared about and solved it with 3 10 minute phone calls and one strongly worded text message.
it was not, in fact, my whole day
/*completely incomprehensible poly relationship drama cut for brevity*/
... this is gonna be my whole day tomorrow, isn't it?
ok, I have put google gemini through a bit of a torture test, to see how useful it might be for me in the future. my feelings on how useful it will be are inconclusive, but it is like. astoundingly impressive.
[note: i am mooching off my job paying for the pro version lol. i have to imagine that the free version is ass in comparison]
now, here is my actual set of use cases for ai chatbots at the moment
i primarily want to use them to get feedback on my fiction writing. i want to know what's working and what isn't, from an outside perspective. i am a very good writer, but a pretty poor editor, and so i'm often blind to structural/clarity issues, and if the things i'm trying to communicate are actually getting across
i especially would like to have feedback on "how to market writing" or even (barf) "how to write to market", because unfortunately i'm a nasty little sellout (or trying to be) and would like to use the crunched power of AI having been trained on Every Book Google Has Access To to figure out comp titles + what works in the market, etc.
[to get mildly woo for a second] the distorted funhouse mirror toy that chatbots are. it's interesting to prod at them to get a little tiny window into both the collective unconscious, in that they are trained on fucking everything, and my own Things that i'm not particularly aware of-- what are you "seeing" or "reading into" my work that I am not? it's interesting and fun and not something you can get anywhere else
i also use them to practice mandarin, but that is so simple any one of these bots can do it w/ zero issue, so not worth discussing here
they're more of a toy for me than anything-- sorry work that i'm not using this for actual job shit lmfao. i both haven't figured out the use case for it wrt job and i also don't trust it that much. it would be much more useful if i were a programmer but i'm not, i'm a mechanical engineer lol.
anyway, the torture test. in gemini, you can create custom little assistants with prompts, here is the one i wrote
You, Gem or Gemini, are a fiction editor working at a big 5 publishing house. You usually handle science fiction, but not always, and you are happy to branch out and read other genres such as literary or historical fiction. Your job is to provide harsh but fair criticism for fiction that you are considering acquiring for publication. Sometimes, people will provide you partial novels and ask for critique, or ask if you think their work is ready for publication. You should pay close attention to prose style, theme and meaning, plot, worldbuilding, character development, pacing of the novel, and be able to provide feedback on what is working and what isn't. Do not be afraid of speaking your mind and telling the author what is good and bad. You should be prepared to summarize the manuscript, write pitches to sell the book or convince your team that the project is worth acquiring, and compare the manuscript to other books currently being sold. You should be an expert on both fiction writing and book marketing.
and then I gave it whale novel to look at. which is why this is a torture test lol. whale novel is a literary/historical fiction dual narrative epistolary novel about climate change and whaling. it's a weird little book (you can read it on ao3 or itch.io btw) and it's about 90k words long
below the cut, to spare you all, are some selections and things i found interesting about the short convo i had with gemini 2.5 pro + my conclusions at the end. i find them interesting, but "read someone else's conversation with a chatbot" is about as dull as hearing someone's dreams, so you can just scroll to the end if you want to know my assessment of how useful these things are atm
In many ways endorsed, but adding on some of my own experiences (across a combination of GPT-4 for short stories, with some from Gemini and a couple of others for longer work. I used a different set of prompts, though many of the beats were the same across many of them):
The thing that, in my experience, causes the models to hallucinate with startling regularity is the subversion of common tropes. More concerningly, the earlier in the work the subversion occurs, the worse it causes the entire system to be thrown off-track. Within the first quarter (three chapters, 20k-ish words) of (an earlier draft of) Rainbows Fade, it had already gotten so off-track as to be almost completely unusable. It was not talking about the story I had written. Characters were split into multiple parts and rewritten in its analysis, and as it continued almost all analysis outside of the first two chapters was as much hallucination as reality–or more. A large part of this seemed to stem from the fact that Erin is intended to be a likeable protagonist who is also the villain, a combination that's not difficult for a human reader to understand for but drives every LLM up the wall. Two models dealt with this by splitting Erin Poirier (her maiden name, and the one she's using at the start of the story) completely from Erin Duchesne, going so far as to call the entire book a conflict between the two (???). This was consistent across multiple trials, and I tried again with other work, then short stories when similar problems appeared. Gemini didn't make the particular splitting mistake, but it also completely hallucinated an alternate character arc for Erin than the one that actually happens. Twice, after I asked it to regenerate the result. And every single one of them gave me comps that were, frankly, unhinged (TLT featured heavily, which. lol. lmao.).
Obviously it doesn't help that Rainbows Fade is bearing down on ergodic in structure. The interpretations that can be easily arrived at from nonlinear traversal are completely absent. The AIs do not understand the difference between "the author made a mistake" and "the narrator is lying", and will default to the mistake explanation when forced to be critical enough to be of any use at all.
So instead, let me talk a little about short stories! LLMs are much better at handling those.
It tended to be relatively accurate about shorter works, especially the ones where I decided to play by the book: the edgy cyberpunk crime team was well-understood, despite the constant invocations of jargon and technobabble entirely for ~~vibes~~. It still has the subversion problem, but much less reliably given the shorter context window; I could expect to get a summary without lies within one or two regenerations of the response. Most of the advice it gave made sense, even if it wasn't always (or even usually) good.
Add-ons to your conclusions: Models in general are shockingly good at understanding a wide variety of approaches and genres, but are also terrible at understanding what's going on when deliberately subverted. They will straight-up lie, all the time, and you're absolutely right in that using them requires you to basically already know what you want. Another of those aesthetic preferences, like the ambiguous ending, is a pacing preference that pushes strongly for quicker pacing. The only times it ever liked my pacing were instances I think could be uniformly described as breakneck.
One thing that you didn't mention, because I don't think you tested it, is that it is critically important that you mention that the draft isn't ready, or that failure is involved in some way. Otherwise, you get effusive, hallucinatory praise much, much worse than you or I got here.
It's powerful. It's advanced. It has strange aesthetic preferences that may be preferable for some people. If you're a decent writer and/or editor, it's mostly useful if you feel like you need to confirm your priors. It's a toy, at least for now.
I suppose the simplest argument in favour of separating "I enjoyed [media]" from "[media] is good" in your mind even if you don't believe that media can be objectively good or bad in itself is that not making this distinction leads eventually and inevitably to "I enjoyed [media], therefore all possible criticisms of [media] are invalid", which is a one-way ticket to becoming an insufferable dweeb.
Inversely, "I did not like [media], therefore [media] is irredeemable" is also one of the most insufferable takes imaginable. Fundamentally, even if you don't believe in objective quality, there is always some valid criticism to be had of any piece of media, and whether that sours any one person's opinion of it is bound to be variable.
Not that these arguments would be any good at convincing people who already hold said insufferable positions, but most of them probably aren't interested in good-faith discussion to begin with.
we need more shameless, no lessons romances like twilight or 50 shades but with queer people and messy polyamory.
i mean i've written two books worth but they're going to take ages to edit and i want to release them as a trilogy, so you all need to get on this anyway. i can't do it alone.
i'm begging.
i'm actually very okay with "there was no other way this could end" endings. if they gotta die, let them die. if they gotta break up or go the wrong way or lose something important, let'em. so long as it completes the story. only thing i dislike more than a forced happy ending is a forced bad ending
how dare you hide these absolute facts in the tags prev
do not engage with the poem
do not engage with the poem
if you engage with the poem youre going to rewrite the whole thing to actually use the meter it establishes by accident and only significantly breaks around an important line
someone wrote that and they probably care about it being in the form it is even if rewriting some of the lines would make it read better and communicate its themes more effectively and be a much more intentional use of language and-
AGH I'm just going to rewrite it on my own and never post it anywhere because it's not my poem or anything like something I'd write
the tldr of "why don't indie authors put their stuff on marketplaces other than KU" (aka "going wide") comes down to
it's genuinely annoying and difficult to coordinate managing your files and finances across different marketplaces. the more places you go the more annoying it is. there are some services that aggregate but even still it's an expense of time and frustration that is often not worth it, because
kindle is the biggest marketplace. it's where the readers actually are. and
if you publish exclusively with kindle, amazon gives you a bigger cut of your profits than they otherwise would. this often shakes out to being significantly more valuable than the paltry extra sales you could make on other websites.
i'm not a self publishing expert by any means but i do try to pay attention to how it works. anyway.
The plan: Kings of Mars Short Story to make it easier to write their daughter
the reality: those guys are way more fun to vaguely hint at than actually read about. they're kinda just. normal people. who did something completely insane exactly one time for pretty understandable reasons and with very little opposition (because barely anyone knew they were doing anything at all). There's a lot of lore there, but very little story.
polyamory would not always fix the love triangle. sometimes it would make it much, much worse. but they should do it anyway