Post AzoISL8EcGcFf1TA0G by impermanen_@zirk.us
(DIR) More posts by impermanen_@zirk.us
(DIR) Post #Azo2QN38urS5lN7vEW by futurebird@sauropods.win
2025-11-01T15:43:34Z
0 likes, 1 repeats
Someone has suggested LLMs are "good for writing summaries" but today I wrote a summary of all 38 chapters in "The Book I Can't Talk About Until it is Done" and this really helped me put them in order better and decide what I need to write next, cut and revise. Just having the summary done wouldn't have done this. And reading the summary wouldn't either. I may never look at the summary again. Making it was the point.
(DIR) Post #Azo2X8GmGaJxwzjE9o by futurebird@sauropods.win
2025-11-01T15:44:50Z
0 likes, 0 repeats
I'm both proud and horrified that the book is long enough that I need to do this. Next I will do a character sheet and describe all of the characters. That should help me check that they make sense as real people among other things.
(DIR) Post #Azo2gFnFf90htKGdiS by lapis@elekk.xyz
2025-11-01T15:46:26Z
0 likes, 0 repeats
@futurebird :sickos: yes... write a long book yes ha ha ha yes!!!(seriously keep at it and good luck)
(DIR) Post #Azo2mpQK6T628Mg5ce by futurebird@sauropods.win
2025-11-01T15:47:39Z
0 likes, 0 repeats
One of the biggest flaws I've noticed from the summary is there aren't enough "ant chapters" VERY excited to fix this.
(DIR) Post #Azo3KRhOoj22Ooh9ZQ by llewelly@sauropods.win
2025-11-01T15:53:43Z
0 likes, 0 repeats
@futurebird I used to have hand-drawn spreadsheets of characters, places, concepts, and so forth, as well as weird diagrams of boxes and arrows, all on graph paper. And all of that for stories I never finished, and for D&D dungeon-mastering prep.
(DIR) Post #Azo3LPWqN6EtxSDZRo by admin@mastodon.slightlycyberpunk.com
2025-11-01T15:49:32Z
0 likes, 0 repeats
@futurebird ...also they aren't actually good at that :)https://www.bbc.com/news/articles/c0m17d8827ko
(DIR) Post #Azo3LQq1VHAw1ECOyu by futurebird@sauropods.win
2025-11-01T15:53:46Z
0 likes, 0 repeats
@admin Well that, and I am not giving the Demon Machine my precious writing. No.
(DIR) Post #Azo4DI66DX0gY0Ot6m by admin@mastodon.slightlycyberpunk.com
2025-11-01T16:03:37Z
0 likes, 0 repeats
@futurebird Yeah, I mean that's probably the more important thing, just a bit of adding insult to injury there.
(DIR) Post #Azo4RVecbt6EeCKOjQ by slotos@toot.community
2025-11-01T16:06:10Z
0 likes, 0 repeats
@futurebird People tend to underestimate the value of the road taken.
(DIR) Post #Azo5MnhcBC1v852wXQ by EriogonumDarwin@mastodon.social
2025-11-01T16:16:24Z
0 likes, 0 repeats
@futurebird This is why I am concerned when students tell me they use LLM to create study guides. The work to create the guide or table IS the studying for me. When I review the guide I recall the logic I used to make it, and THAT is the studying, not memorizing another flat pattern of words.
(DIR) Post #Azo5fhr299wiUgT0vw by CleyFaye@mastodon.top
2025-11-01T16:19:57Z
0 likes, 0 repeats
@futurebird And that's assuming they actually *are* good at it. Unfortunately, determining what is meaningful and what isn't is not something an LLM can do, so… I suppose people finding automated summaries helpful either work with very small source material, or took it from granted and never looked at the source material.
(DIR) Post #Azo63L34ad3OiGMoam by mayintoronto@beige.party
2025-11-01T16:24:13Z
0 likes, 0 repeats
@futurebird WAIT. You're writing a book AND teaching?! Where do you find the time?
(DIR) Post #Azo6RGVj7499aaB8e8 by noodlemaz@med-mastodon.com
2025-11-01T16:28:31Z
0 likes, 0 repeats
@futurebird they can't summarise. To summarise, you need context. You need to know more than just the words in the thing.They don't.They don't know anything. It's just an assessment of the document on its own, nothing more. I wish people would stop thinking they did this. Not to mention the random bullshit injections.
(DIR) Post #Azo72lboSBUOkjuKQK by deborahh@cosocial.ca
2025-11-01T16:35:21Z
0 likes, 0 repeats
@futurebird @mayintoronto 💯I do wonder how the world will be, when these generations that never learned to read and write on their own are in charge. 😱
(DIR) Post #Azo89xmng3JyV82g9Q by jwcph@helvede.net
2025-11-01T16:47:49Z
0 likes, 0 repeats
@futurebird Also, LLMs don't summarize - they shorten. https://ea.rna.nl/2024/05/27/when-chatgpt-summarises-it-actually-does-nothing-of-the-kind/
(DIR) Post #AzoBtmUKJZjGe0wNKS by sashin@veganism.social
2025-11-01T17:29:44Z
0 likes, 0 repeats
@futurebird oh no, that's make it longer!!!
(DIR) Post #AzoCs4NUvhJxVIUheK by Sylvhem@eldritch.cafe
2025-11-01T17:40:34Z
0 likes, 0 repeats
@futurebird People have also pointed out that LLM are not actually that great at summarizing things. They are great at giving you a summary for something that’s already been summarized, i.e. scientific papers and topics covered by Wikipedia.
(DIR) Post #AzoEDkUHf9Ms8hZNVA by maxleibman@beige.party
2025-11-01T17:55:43Z
0 likes, 0 repeats
@futurebird This is how note-taking always worked for me in school. I didn’t “study” the notes. Figuring out what to jot down was processing, and was usually sufficient for me to be able write or test on whatever was covered. The only studying I would ever do with my notes was to re-write and clean them up when needed, but again I rarely referred back to even the cleaned-up ones. It’s all about the act.
(DIR) Post #AzoEU6L8BSY0Xx5O0u by crcollins@writing.exchange
2025-11-01T17:58:41Z
0 likes, 0 repeats
@futurebirdIt helped me, while in college, to write out combined summaries of my lecture notes and assigned readings.
(DIR) Post #AzoFQUiLia3owFn2mm by StephanieMoore@mastodon.online
2025-11-01T18:09:12Z
0 likes, 0 repeats
@futurebird Summarizing is a great strategy with several good studies behind it (generative learning theory). It’s a shame it’s one of the most susceptible to LLMs.
(DIR) Post #AzoHNt4Ytje7WY43HM by Delphi@mastodon.scot
2025-11-01T18:31:11Z
0 likes, 0 repeats
@futurebird @admin 👍
(DIR) Post #AzoISL8EcGcFf1TA0G by impermanen_@zirk.us
2025-11-01T18:43:11Z
0 likes, 0 repeats
@futurebird this is how it is writing software, too. The process of feeling rough edges, stuff that isn’t clear, things that are out of balance or unwieldy… all that felt sense is what drives the gradual refactor to good architecture, a system capable of being maintained. You don’t get any of that if you’re vibe coding with an LLM. I think you must become numb and eventually quite helpless, even if you receive some applause for a while at being quick.
(DIR) Post #AzoPKIPjXJNYatlXGa by raymaccarthy@mastodon.ie
2025-11-01T20:00:09Z
0 likes, 0 repeats
@futurebird Also nearly half of AI summaries are wrong and without reading the entire source you'd not know which half.
(DIR) Post #AzoQRp0cqZodfQAr9E by ASprinkleofSage@mastodon.social
2025-11-01T20:12:41Z
0 likes, 0 repeats
@futurebird pretty sure that whole 'good at summaries' line is to get people to feed it more content
(DIR) Post #AzoWxPsgPKguj1uLKa by nazokiyoubinbou@urusai.social
2025-11-01T21:25:42Z
0 likes, 0 repeats
@futurebird I never understood how people could think they're good at summaries. Putting aside that they are, in practice, demonstrably not, for stuff like that you would be exceeding the context window anyway...
(DIR) Post #AzoYWuVIe0otn2SWdk by htpcnz@mastodon.social
2025-11-01T21:43:17Z
0 likes, 0 repeats
@futurebird yeah gave copilot a chance to identify a plant as a test and it flew right at a mountain still believing it is only a molehill... Finally even it agreed better approach is to parachute out and ask someone from the local garden centre...
(DIR) Post #Azol4M3yOpkKGmOKA4 by CStamp@mastodon.social
2025-11-02T00:03:45Z
0 likes, 0 repeats
@futurebird I think if you want to know the material vs present the material, it's important to do it yourself.