Carbonado
60 posts
Carbonado
@carbonadoks
Living in the past, seeing into the future. more active on https://t.co/nOmU4Be6jQ
Katılım Eylül 2024
259 Takip Edilen22 Takipçiler
@eriskiiii @norvid_studies has written a good argument about pangram
bsky.app/profile/norvid…
like the whole granta situation. the short stories are clearly ai written and the authors lied about their authorship and i think it is good that pangram detects these as ai written.
English
@eriskiiii Can you give some examples where pangram misclassified a human text as ai? I really think pangrams strength is its false positive rate. I read you complains about pangram on bsky but i think its really rare that pangram classifies human text as ai.
x.com/carbonadoks/st…
Carbonado@carbonadoks
@norvid_studies @turtlelambvase @lu_sichu @max_spero_ @voooooogel i just tested it twice and typed just random stuff into pangram and both times my writing was classified as human written. i really want to see some real human written short fictions classified as ai. like maybe make a competition how easy it is to write an ai short story.
English
Pangram has a horrible false positive rate. It is a fundamentally antisocial technology. Who are they to dictate what writing is 'good' and 'bad'? What editing tools are 'okay' or not? It should not exist. It's also trivial to get around by having the model iterate on results.
Simon Grimm@Simon__Grimm
Pangram is one of the most pro-social, AI-centered technologies of the past year and a great example of defensive acceleration. If you don't use the plugin, do. It automatically scans Twitter and LinkedIn feeds, see below. It also flags AI-assisted Substack posts.
English
@norvid_studies @turtlelambvase @lu_sichu @max_spero_ @voooooogel x.com/voooooogel/sta…
you can write like an ai but i think writing a short story like ai is so much harder.
thebes@voooooogel
i'm a @pangramlabs defender, but it turns out it *is* possible to get a false positive from them if you really try to write like claude 😀
English
@norvid_studies @turtlelambvase @lu_sichu @max_spero_ @voooooogel i just tested it twice and typed just random stuff into pangram and both times my writing was classified as human written. i really want to see some real human written short fictions classified as ai. like maybe make a competition how easy it is to write an ai short story.
English
does anyone think that data scrapers care about the color of your skin(how the fuck would they know anyway) this is sheer garbage. when they collect writing data, they are stealing library genesis and every fanfiction website out there that have an exposed api for downloading en masse.
"unusual writing process" honestly if they come out and admit they are doing ai writing experiments(like @gwern) that would be much more impressive. it takes skills to edit and prompt the llms not to sound like nails being scraped on a blackboard right now
Joe Weisenthal@TheStalwart
There seems to be an accusation here that @pangramlabs only used work that comes from the “dominant culture” and therefore it’s unreliable at measuring text used in the context of this prize, which aims to reward underserved communities.
English
@norvid_studies tbh i did not memorize anything for tetris i just try to play as fast possible. but i think i subconsciously memorized a lot of patterns in tetris and it just really fun to execute something really fast. thats the same for speed cubing for me you just pattern match and get really
English
tetris is more like rubiks cubing in terms of finite set of branches memorized in advance? TIL
Irojeh@irojeh1
@MoonL88537 @kosenjuu @norvid_studies The current state of PCs would I think be akin to overfitting on the training set. There are only so many unique situations, few enough that we can find the optimal solution for any given state. The better players essentially just memorize the optimal decision tree for...
English
English
@carbonadoks @FleischmanMena @norvid_studies Think that's a diff player, my tetrio name is Dreameville but my username for everything else is JimothyJimothy. Here's my jstris and tetrio graph. Tbh the sprint grinding all kinda blurs together so I couldn't tell you
English
@norvid_studies Used to do speedruns. A valuable community norm was “if you’ve got a better time from an unknown legend, feel free to post it, otherwise I’m gonna keep claiming WR till you do.”
Shifts the burden back to claimants, who now lack WR, AND are shown to be failed web researchers.
English
@FleischmanMena @Algon_33 @norvid_studies Any more thoughts on centauring? I thought of doing this as well just to make my writing more clearer but it produced only slop.
English
@Algon_33 @norvid_studies Collaborative writing with models to produce downstream artifacts.
I stopped doing it outside self-fact-checking as I felt it diluted my voice too much, I don't need a second moron to be a full one
English
Carbonado retweetledi
@medjedowo @FleischmanMena @norvid_studies Heh. That said, I think I would recommend getting close to the frontiers of anything somewhat competitive, especially if that requires being part of a community, so you can recognize what it's like to be at the edge of human ability, and to show you that you can reach such edges.
English
@irojeh1 @FleischmanMena @norvid_studies I used to play a lot more but it's pretty sad to see your imrpovement stats in jstris.
English
@irojeh1 @FleischmanMena @norvid_studies Damn that's really sick. Did you have any plateaus reaching 22s Sprints or was it a gradual improvement until this level? I think i got hard stuck at like ~35s and my pb is 30.9s
English
Carbonado retweetledi
New day, new post. Today is about why having a superhuman AI available at all times might not change your life much.
English
thread of claude drawings
thebes@voooooogel
req'd by a moot: happy claude, free from trolley problems :-)
English
Carbonado retweetledi
Carbonado retweetledi
