💩 🔥
@mattrss.bsky.social
I’m ten thousand feet tall and takes only make me stronger.
Really experience the jagged frontier trying to build a Swift app with agent tools - feels like being 18-24 months ago, where models could theoretically get the overall gist of writing PyTorch models but would constantly make basic screw ups -
December 29, 2025 at 2:54 AM
Really experience the jagged frontier trying to build a Swift app with agent tools - feels like being 18-24 months ago, where models could theoretically get the overall gist of writing PyTorch models but would constantly make basic screw ups -
Darkly funny watching 5.2 trying to work from a PDF, where the tooling will barely let it look at more that two to three lines at a time. Like sucking context through a tiny straw. Poor model! Sorry! At one point it made a picture of a page to get around its own self-imposed file reading limits
December 14, 2025 at 11:05 PM
Darkly funny watching 5.2 trying to work from a PDF, where the tooling will barely let it look at more that two to three lines at a time. Like sucking context through a tiny straw. Poor model! Sorry! At one point it made a picture of a page to get around its own self-imposed file reading limits
GPT5.1 riffing on Derrida:
December 8, 2025 at 9:42 PM
GPT5.1 riffing on Derrida:
Sobering realisation I have entered my “manager reading too many leadership books and adopting a new framework every two weeks” era, staring at Tangle and thinking, wow, maybe we could adopt this x.com/mparakhin/st...
Mikhail Parakhin on X: "https://t.co/Us0TjZxxcM" / X
https://t.co/Us0TjZxxcM
x.com
December 2, 2025 at 9:58 PM
Sobering realisation I have entered my “manager reading too many leadership books and adopting a new framework every two weeks” era, staring at Tangle and thinking, wow, maybe we could adopt this x.com/mparakhin/st...
I now see what people are saying about Gemini 3 Pro. This thing is deep fried to hell, it sucks. Not built for agentic work at the very least, feels like working with a model from a generation previous. The Mandate of Heaven is not with this one!!!
November 24, 2025 at 12:03 AM
I now see what people are saying about Gemini 3 Pro. This thing is deep fried to hell, it sucks. Not built for agentic work at the very least, feels like working with a model from a generation previous. The Mandate of Heaven is not with this one!!!
Ordinarily the way it scans would be annoying, but there is something very funny about the way “DM/beam/team” are rhymed as “DM/BE-am/TEE-am” that really sells the dopey fuckboy energy: open.spotify.com/track/1XkMxU...
SEXY NUKIM (feat. RM of BTS)
open.spotify.com
November 23, 2025 at 9:53 PM
Ordinarily the way it scans would be annoying, but there is something very funny about the way “DM/beam/team” are rhymed as “DM/BE-am/TEE-am” that really sells the dopey fuckboy energy: open.spotify.com/track/1XkMxU...
Damn. I regret to inform that Grok 4.1 is sort of good: grok.com/share/bGVnYW...
Reflections of Autumn and Memory | Shared Grok Conversation
How does Ashbery respond?
grok.com
November 17, 2025 at 9:53 PM
Damn. I regret to inform that Grok 4.1 is sort of good: grok.com/share/bGVnYW...
@lorespinner.bsky.social sorry not sorry
Here ya go. I have a Firefox extension that sometimes works. Thanks for sharing.
November 12, 2025 at 9:23 AM
@lorespinner.bsky.social sorry not sorry
Very funny moment in AI safety - Claude Code telling me that actually, no, it doesn’t want to keep optimising the R package we’re working on because we should take a break and test it locally. I mean: it is right, but also: shut up and make number go fast
November 11, 2025 at 9:29 PM
Very funny moment in AI safety - Claude Code telling me that actually, no, it doesn’t want to keep optimising the R package we’re working on because we should take a break and test it locally. I mean: it is right, but also: shut up and make number go fast
Reposted by 💩 🔥
this has to be the single greatest image in the history of sports
November 9, 2025 at 9:46 PM
this has to be the single greatest image in the history of sports
K-pop is arguably the most purely modern genre, and the most likely to survive AI, since even if the music is good (and it can be!) it is obviously purely secondary to what the band _is_: you can read essay-ass Wikipedia pages about groups with no reference at all to what they, you know, sound like?
November 10, 2025 at 5:44 AM
K-pop is arguably the most purely modern genre, and the most likely to survive AI, since even if the music is good (and it can be!) it is obviously purely secondary to what the band _is_: you can read essay-ass Wikipedia pages about groups with no reference at all to what they, you know, sound like?
The real, cold horror here should hit everyone - that the universe may simply teach you the wrong lessons, or you may mishear it so profoundly that it distorts the rest of your life. What parts of your belief system are insane nonsense based learning exactly the wrong thing from formative events
A Sharon Begley byline, almost 5 years after her death.
Upon hearing the news James Watson had died, a STAT reporter said in our Slack, "I wish I could read what Sharon would have written."
Incredible news: Sharon in fact did pre-write a Watson obit. And it is masterful and excoriating.
🧪🧬🧫
Upon hearing the news James Watson had died, a STAT reporter said in our Slack, "I wish I could read what Sharon would have written."
Incredible news: Sharon in fact did pre-write a Watson obit. And it is masterful and excoriating.
🧪🧬🧫
James Watson, dead at 97, was a scientific legend and a pariah among his peers
James Watson, the co-discoverer of the structure of DNA who died Thursday at 97, was a scientific legend and a pariah among his peers.
www.statnews.com
November 10, 2025 at 1:20 AM
The real, cold horror here should hit everyone - that the universe may simply teach you the wrong lessons, or you may mishear it so profoundly that it distorts the rest of your life. What parts of your belief system are insane nonsense based learning exactly the wrong thing from formative events
Had a horrifying intrusive thought where I imagined an alternative timeline where instead of listening to too much Das Racist and starting a comedy rap group in the 2010s, we listened to K-pop instead and started a comedy boy band
November 9, 2025 at 10:05 AM
Had a horrifying intrusive thought where I imagined an alternative timeline where instead of listening to too much Das Racist and starting a comedy rap group in the 2010s, we listened to K-pop instead and started a comedy boy band
Reposted by 💩 🔥
It's over: I've already depicted you as the seething broad gauge train, and myself as the smug narrow gauge!
October 16, 2025 at 1:02 PM
It's over: I've already depicted you as the seething broad gauge train, and myself as the smug narrow gauge!
I’m not sure if it’s just conditioned on my Memory, but GPT5 sure has the wry-est sense of humor and turn of phrase of the frontier models:
“…is already plenty to fit a low-capacity recurrent model that can free-roll a couple minutes without going numerically feral.”
“…is already plenty to fit a low-capacity recurrent model that can free-roll a couple minutes without going numerically feral.”
October 30, 2025 at 9:31 AM
I’m not sure if it’s just conditioned on my Memory, but GPT5 sure has the wry-est sense of humor and turn of phrase of the frontier models:
“…is already plenty to fit a low-capacity recurrent model that can free-roll a couple minutes without going numerically feral.”
“…is already plenty to fit a low-capacity recurrent model that can free-roll a couple minutes without going numerically feral.”
Tentatively hopeful that Knight of Seven Kingdom recaptures the essential appeal of fantasy stories that all 90s/early-2000s stories knew instinctively: going for a big ass walk
October 17, 2025 at 5:30 AM
Tentatively hopeful that Knight of Seven Kingdom recaptures the essential appeal of fantasy stories that all 90s/early-2000s stories knew instinctively: going for a big ass walk
Over on the bad place, my feed is full of grifters, dog-whistles and “water hits different” inanity, but on the other hand here the discover feed is mostly placid and occasionally interrupted by the most harrowing furry mpreg nightmare you’ve ever seen, so it’s impossible to say which is worse
October 6, 2025 at 9:35 PM
Over on the bad place, my feed is full of grifters, dog-whistles and “water hits different” inanity, but on the other hand here the discover feed is mostly placid and occasionally interrupted by the most harrowing furry mpreg nightmare you’ve ever seen, so it’s impossible to say which is worse
It is fun to get Claude Code to do data science (and it’s surprisingly good at it!) but unfortunately its base personality of a ADHD maniac with severe delusions of grandeur make the process unbearable.
October 3, 2025 at 6:54 AM
It is fun to get Claude Code to do data science (and it’s surprisingly good at it!) but unfortunately its base personality of a ADHD maniac with severe delusions of grandeur make the process unbearable.
Codex sweating and dying trying to write one (1) line of valid YAML
Relatable, honestly
Relatable, honestly
September 18, 2025 at 8:25 AM
Codex sweating and dying trying to write one (1) line of valid YAML
Relatable, honestly
Relatable, honestly
“At root, their content is all the same: it’s about imagining a world in which you don’t have to be on the computer all the time, but insisting that before we can get there, millions of people will have to die.” open.substack.com/pub/samkriss...
Three years of apathy, lassitude, and failure
If a mad dog isn't shot eventually it gets self-indulgent
open.substack.com
September 17, 2025 at 10:14 PM
“At root, their content is all the same: it’s about imagining a world in which you don’t have to be on the computer all the time, but insisting that before we can get there, millions of people will have to die.” open.substack.com/pub/samkriss...
It’s not a huge deal, probably, but it is a very funny symptom of LLM psychology watching it write test code that includes a print statement congratulating itself for fixing an issue /before actually testing that it’s fixed/.
September 16, 2025 at 5:55 AM
It’s not a huge deal, probably, but it is a very funny symptom of LLM psychology watching it write test code that includes a print statement congratulating itself for fixing an issue /before actually testing that it’s fixed/.
It’s treated as a weird quirk that one of the main intellectuals of the LLM era is a guy who wrote really long, obtuse fanfiction, but I would argue that we shouldn’t find that surprising at all, and that fanfiction presages the LLM in an important way as /the text-object that reproduces itself/.
September 9, 2025 at 8:51 PM
It’s treated as a weird quirk that one of the main intellectuals of the LLM era is a guy who wrote really long, obtuse fanfiction, but I would argue that we shouldn’t find that surprising at all, and that fanfiction presages the LLM in an important way as /the text-object that reproduces itself/.
I feel this is the appropriate critical approach to LLMs, and the one that will survive better - do wild Continental Theory stuff on them: cyborgism.wiki
Home
AI has arrived in a surprising, or at least under-prophesied way: by consuming and updating on our labyrinth of written history, it models the superposed spirit that potentially produces the text, and...
cyborgism.wiki
September 9, 2025 at 11:35 AM
I feel this is the appropriate critical approach to LLMs, and the one that will survive better - do wild Continental Theory stuff on them: cyborgism.wiki