⁂ L. Rhodes

Tinkerer. Reverse reverse centaur. Fix your hearts or die.

pronouns: he/him/they
site: https://lrhodes.net/
tech: https://destructured.net/

Public Key

npub1e5z8avwreewgmjhugmusjpddnyrgjgnarahvfawh9m566pezkyysu2p90v

NIP-05 Address

lrhodes_at_merveilles.town@momostr.pink

Profile Code

nprofile1qqsv6pr7k8puuhydet7yd7gfqkkejp5fyf737mky7htja6ddqu3tzzgprpmhxue69uhhyetvv9ujumt0d4hhxarj9ecxjmntqywhwumn8ghj7mn0wd68yttsw43zuam9d3kx7unyv4ezumn9wsgdaae3

Publishing to

relay.momostr.pink

nostr-pub.wellorder.net

yabu.me

Author Public Key

npub1e5z8avwreewgmjhugmusjpddnyrgjgnarahvfawh9m566pezkyysu2p90v

Show more details

Last Notes

2026-10-30 22:53:35 UTC

- reply

Personal computers were sold to us as this incredibly empowering technology, and now they're trying to claw that power back so that they can meter it and place limits on what can be done with it.

2026-10-30 22:45:36 UTC

- reply

Let me be clearer about what that would entail. You want to set a reminder on your computer? Sam Altman wants that interaction to go through his data center. Not just the text and alert time, but the processing required to handle reminding as an activity. That's a tiny task relative to the power of any modern computer, but he wants to strip home computers down to the point where they can't do anything without routing the task through his company first. The goal is to capture computing itself.

2026-10-30 22:27:26 UTC

- reply

This is barely even news, except that he's being so open about it now. It's something I've pointed to for a while now. If you step back and take a broad view of the #AI industry, a clear trajectory of what they've been doing is toward a situation where ALL personal computing takes place in their hyperscale data centers, and we all just pay a subscription fee to connect via low-powered terminals. "Having RAM at home is wasteful" is an argument for deep dependence on him for ALL of our computing.

2026-10-30 22:14:17 UTC

Dude wants to centralize memory. https://www.ign.com/articles/openai-ceo-sam-altman-thinks-its-terrible-for-you-to-have-ram-in-your-gaming-pc

2026-10-27 15:18:29 UTC

- reply

It's a good panda.

2026-10-27 15:17:46 UTC

A big welcome to the Brazilian people trying out the fediverse for the first time. We have a few important rules of etiquette here, but you're already in compliance with the first one: Block all fascists.

2026-10-26 23:06:13 UTC

- reply

I don't even remember why I decided against Proton, but I'm grateful to past me for dodging that bullet.

2026-10-26 21:03:37 UTC

AI policy
https://assets.merveilles.town/media_attachments/files/117/509/001/133/396/987/original/4391cc4d4f9bc497.jpg

2026-10-26 17:04:08 UTC

If you're new to the fediverse, and someone asks to see your library card, it's not a form of age verification, we just get really excited about libraries.

2026-10-26 15:00:53 UTC

- reply

It must be very disconcerting to have the computer consistently tell you you're right and make good points, and then log into a forum and find that the computer is suddenly telling you that you're wrong and that those points actually aren't good at all.

2026-10-26 14:57:37 UTC

- reply

"If people here continue to oppose LLM use, I will have no choice but to retreat into solipsism."
¯\_(ツ)_/¯

2026-10-24 21:49:07 UTC

I wouldn't say that a person who opposes AI on the premise that it might attain consciousness and destroy humanity really qualifies as an "AI skeptic."

2026-10-23 13:03:51 UTC

It seems to me that the last several years have demonstrated the shortcomings of FOSS as a set of organizing principles for computing, so I've developed a set of principles meant, not as a replacement, but as a corrective, and written them out as a sort of manifesto. Check it out here: https://destructured.net/casa
Feedback welcome. Much thanks to @npub1l2s…vp8m and @npub13e0…qty3 and @npub1z8r…mrq6 for theirs.

2026-10-22 22:04:19 UTC

- reply

Probably not, but moderation and community maintenance are burdens for every service on the fediverse, not just image-sharing services like Pixelfed, so what I'm talking about is really added burden on top of that.
Just anecdotally, Merveilles was grinding its gears recently, resulting in a ton of errors. The cause, it turned out, was several terabytes of storage clogging up the backend. I imagine that racks up pretty quickly when every post is a hi-res image or video.

2026-10-22 18:23:42 UTC

- reply

I was thinking it might be cool to have an ActivityPub service that was all about lo-fi image posting, with conversion to <100kb format happening when you upload. Maybe it even provides a choice of "filters" so you can upload the pic with different types of compression, dithering, minimal color schemes, whatever. Could be a fork of Pixelfed, I suppose, but given Wake's point about its PHP foundation, it might be better to start from scratch.

2026-10-22 18:10:33 UTC

- reply

And, as it turns out, the AI didn't really "discover" the exploit that it may have helped publicize: https://infosec.exchange/@mayahustle/117485672263060599
Everyone, from mathematicians to infosec researchers, needs to internalize the lesson that LLMs are inherently leaky, and not to be trusted with anything you want to publish on your own terms.
#note1yvt…9z69

2026-10-22 17:34:51 UTC

- reply

I think the bigger obstacle for building viable ActivityPub replacements for the corporate photo sharing apps may be storage costs. It's a wall I keep expecting Pixelfed instances to hit.

2026-10-22 17:32:16 UTC

Chainguard, which basically leaked PHP2Shell ahead of the patch, says they figured out the details by querying "a frontier model" with the prompt "find bad php vulnerability": https://www.linkedin.com/feed/update/urn:li:activity:7518756994284077056/
That could happen a few ways, but the likeliest may be that they used the same model that found the vulnerability in the first place, and it just… divulged what it had found. Which would mean cloud AI can't be used to test code without also potentially publicizing vulns before they're patched.

2026-10-22 14:19:37 UTC

- reply

sysadmin: hey, chat, tell me about this vulnerability your company helped patch.
chatbot: sure, i'd love to help with that!
chatbot <sending local data to a datacenter>: basically, it was a potential backdoor
chatbot <running sudo commands>: … that would allow remote users to inject admin commands
chatbot <updating to vibecoded OS version>: … which could allow them to take control of your entire system.
sysadmin: oh wow, that sounds bad.
sysadmin: …
sysadmin: waitaminute

2026-10-22 14:00:23 UTC

- reply

FOSS orgs: hey, so uh, this probably wasn't going to be a problem before, but now that it's known, there's a very old vulnerability that everyone needs to patch it right away.
crowd: BOO, YOU SUCK, GET WRECKED
FOSS orgs: sigh. okay.
FOSS orgs: btw, this vulnerability was discovered by an ai company and they contributed to the patch.
crowd: OMG, SO BASED, OUR HEROES
AI corp <ralph wiggum voice>: i'm helping

2026-10-22 13:50:05 UTC

AI corps: hey, thanks for the software!
FOSS orgs: oh. cool? think you could kick a little help our way?
AI corps: oh, totally.
FOSS orgs: awesome, because we could really use—
AI corps: here you go!
FOSS orgs: oh, okay. whatever you can spare is good.
AI corps: it's a previously unexploitable vulnerability that you have to fix immediately or the internet will shatter into a million pieces.
FOSS orgs: …
FOSS orgs: WHAT

2026-10-21 21:28:39 UTC

- reply

The p in dansup stands for php.

2026-10-18 16:55:38 UTC

When your friends who only use corporate social media platforms ask what the fediverse is good for, tell them that, if nothing else, it's a good place to find out what legitimate speech the corporation is quietly censoring and suppressing on the platform they use.

2026-10-16 02:53:16 UTC

One thing I'm only just now appreciating is how the shift to "agentic AI" serves to undermine any privacy benefits you might get from using a local model. Maybe the model isn't sending every prompt back to a data center, but if you're handing tasks over to agents, then they're likely out there leaving a trail of data all across the internet — and maybe with a preference for sites that filter back to the model vendor.

2026-10-14 16:49:53 UTC

- reply

It's a phenomenon I'm seeing a lot in FOSS circles lately: LLMs amplify a benevolent dictators ability to handle things on their own, which feeds into their need for control. That can manifest as anything from unwillingness to share work, to outright messages of hostility to potential recruits.

2026-10-13 18:27:43 UTC

- reply

"Grandma, how did people keep refrigerators from leaking sensitive photos when you were my age?"

2026-10-13 18:20:51 UTC

You almost certainly have enough cameras and microphones in your home. It's highly unlikely that adding more will improve your life in any way.

2026-10-12 14:40:08 UTC

I might be more inclined to harden my password game if the weak point in every single breach I've ever been exposed in wasn't the company on the other end of my login.

2026-10-11 14:40:53 UTC

- reply

It would fit there, too, yeah.

2026-10-11 12:37:00 UTC

Feels like my fellow weirdos on fedi would make good use of an #EternalFamily hashtag for trading recommendations for things to watch.

2026-10-09 13:43:09 UTC

- reply

My suspicion is that both this and "pacing the frontier" mark a shift from training major releases to incrementally training an in-production model so that it's constantly ingesting data as it operates. Banning abusive behavior allows the system to rapidly convert usable information (e.g. current events, marketable data about users, research approaches) into potential output with less risk that the model will phrase that output in user-hostile terms.

2026-10-08 23:32:41 UTC

Kevin Roose, the guy who gave Claude his credit card info to spin up the Forkiverse, has joined the staff at The Atlantic: https://www.theatlantic.com/press-releases/2026/09/the-atlantic-is-announcing-seven-new-roles-in-the-newsroom/688820/
He and former Hard Fork cohost/Forkiverse admin Casey Newton have an upcoming NPR podcast that bills itself as "the official podcast of the singularity": https://machinegods.fm/
In case you're wondering why they don't have time/energy/interest for maintaining the community they started.

2026-10-08 19:49:40 UTC

- reply

If anyone else is curious or looking, Selfoss is the only self-hosted RSS aggregator I know of that has both post filters and a no-tolerance policy with respect to AI contributions: https://selfoss.aditu.de/

2026-10-08 19:35:07 UTC

Who, me changing RSS aggregators yet again?

2026-10-08 18:36:35 UTC

Seeing techbros talk about how they use AI to automate basic interactions with their families really tells you the sort of people we're dealing with here.

2026-10-08 17:50:35 UTC

- reply

LOL when you can't even get ANNUALIZED revenue as high as you need it.

2026-10-06 20:03:05 UTC

"If AI had existed in the early aughts, we could not have accelerated Zotero's conception, because we did not know exactly what we wanted, and so could not have written coherent prompts for an LLM. Instead, it took a great deal of time and collaboration to develop a clear vision for what Zotero should be. But that slow formation led to software that was durable rather than ephemeral, with a strong foundation that could be built upon." https://newsletter.dancohen.org/archive/the-slow-formation-of-durable-software/

2026-10-06 16:58:37 UTC

Modern frontend design seems heavily influenced by the belief that no one actually wants smaller phones.

2026-10-06 16:40:51 UTC

- reply

Idea: A microwave with a plus button for typing in the pauses.

2026-10-06 13:23:34 UTC

I haven't tried this yet, and I may wait to make sure there are no catastrophic adverse effects, but https://arstechnica.com/apple/2026/10/command-line-tool-quickly-removes-apple-intelligence-from-macos-27/ 👀

2026-10-02 15:49:32 UTC

- reply

They made it sound like Roose paid for it, but I'm not sure we ought to believe that. My guess is that they expensed it.

2026-10-02 15:24:59 UTC

- reply

It was a marketing experiment: Can you use ActivityPub to market a podcast? The only problem being that they presented it as an experiment in building a better social media. Which left a lot of people imagining that their participation mattered as a part of a community, when mostly it was a data point for the success of the show. (And, if their cancellation is any indication, the Times didn't see the Mastodon community as a terribly compelling argument for the success of the show.)

2026-10-01 14:40:32 UTC

- reply

"Well, other technologies have similar costs, and we use those regularly."
There's a difference in scale that matters when we talk about AI, but yes, the general point is valid, and I'd love to spend more time talking about how much damage we're doing with casual use of those other technologies, but right now the normalization of AI use feels like the more acute threat.

2026-10-01 14:39:26 UTC

- reply

"Oh, but AI will become less socially, politically, psychologically and environmentally destructive as the tech becomes better."
Great. When that happens, the range of acceptable uses will expand accordingly. But using it as though it were not costing us dearly now is building up debts that we may never be able to pay.

2026-10-01 14:34:45 UTC

Over and over again, I keep hearing and reading things by people seeking "nuance" in how they approach AI, and the best I can do for nuance is:
If a technology entails huge social, political, psychological and environmental costs, then you should only use it to perform tasks that are worth those costs.
None of the uses people are currently making of AI are worth the actual (as opposed to monetary) costs. Any "nuance" beyond that feels like rationalization.

2026-10-01 12:07:01 UTC

- reply

What about this? https://github.com/hundredrabbits/awesome-uxn

2026-09-29 17:29:28 UTC

This won't be news to some of you, but Colorverse feels like a pretty great daily #fountainPen ink. Great colors (I'm trying out Coffee), quick drying, no smearing or perceptible bleed on Leuchtturm paper. This might be my new go-to.

2026-09-29 16:49:16 UTC

"A new rule from the Office of Management and Budget would prohibit anyone from using federal grant money for voter registration drives." Including efforts that make voting accessible to disabled Americans. https://www.propublica.org/article/disabled-voters-nonprofit-funding-cuts-omb
(To be clear, this likely won't affect the midterms, but it could limit enfranchisement in the 2028 elections.)
#politics #USA #votingRights

2026-09-28 15:55:44 UTC

Here's an interesting paper that breaks sustainable computing down into five principles:
1. minimization of unnecessary abstraction layers;
2. reduction and exposure of background processes;
3. support for attentional continuity;
4. prioritizing clarity over feature expansion;
5. enhancing predictability.
https://ijecs.in/index.php/ijecs/article/view/5492

2026-09-26 13:02:23 UTC

- reply

Without NGO watchdogs and the press, we might not be seeing any action here at all. The state seems not to have even noticed 62 cruise ship gas turbines until they were pointed out to them.
The clear evidence is that hyperscale datacenter operators cannot be trusted to stay within the even their own advertised limits, and state agencies cannot be relied upon to hold them to those limits once they're operational. The best option for communitues is to stop these projects before they break ground.

2026-09-26 12:50:39 UTC

Better than no fine at all, I guess: https://www.theguardian.com/environment/2026/sep/25/new-jersey-fines-datacenter-unpermitted-generators but prior reporting says these are Bergen gas turbines: https://www.theguardian.com/environment/2026/aug/27/new-jersey-vineland-datacenter-permits which range from 1.4MW–11.8MW/engine: https://www.bergenengines.com/on-land/ and going by these industry numbers: https://uspeglobal.com/articles/gas-turbine-epc-costs-2026/ that puts DataOne spending $97.7M–$1.6B on its 62 temporary turbines, so losing $1.1M is like losing use of one turbine temporarily, and they don't even have to stop for another 45 days.

2026-09-25 13:32:59 UTC

- reply

If holograms helped target online ads, I bet we'd have this figured out by now.

2026-09-25 13:21:06 UTC

Whatever happened to holograms? Younger me thought they'd be everywhere by now.

2026-09-24 21:25:45 UTC

- reply

"We can now make it possible, because of AI, that everybody can take advantage of this computer, use it to its limit without having to speak a new language — Fortran, Pascal, C++, Rust, CUDA, every one of those languages."
Nope nope nope. Programming languages translate instructions into code, but just as important is they way they teach the programmer to think like a computer. Building software without understanding how a computer operates is an invitation to disasters you can't anticipate.

2026-09-24 21:13:59 UTC

This is a write-up of an interview: https://www.theverge.com/tech/1000140/jensen-huang-nvidia-ai-energy-climate-change-supervillain and you could just go read the interview in full: https://www.nytimes.com/2026/09/23/opinion/ezra-klein-podcast-jensen-huang.html but what The Verge piece is getting at — rightly, and with valuable context — is that #AI boosters like Huang see sustainability initiatives as an impediment to AI growth, and will push off indefinitely any efforts to place limits on their access to cheap energy. "AI will eventually solve this problem" is just kicking the can down the road.

2026-09-24 15:00:54 UTC

- reply

"AI agents" can't do much without access. To wring value from them, people grant them access — to social media accounts, to credit card info, to school, work, insurance, etc. That gives the AI vendor access to troves of data they couldn't have gotten at otherwise, which can then be used to build stronger, more comprehensive, more intrusive profiles, not just about the AI user, but also anyone else who appears in their data.
They're potentially the most effective surveillance machines ever made.

2026-09-24 14:48:54 UTC

Wild thread, but for my money, the most interesting part is this section: https://neuromatch.social/@jonny/117325356932250166 which highlights how Meta Muse maps in granular detail your relationships to other people. This, of course, carries over Facebook's traditional modus operandi, but having an AI agent draw the connections is like pumping social graph full of steroids. Imagine what sort of inferences it can draw when you're, say, giving it bank access so it can cancel your forgotten subscriptions. #AI

2026-09-24 14:26:40 UTC

Giving myself admin privileges over library folders at the root, usr and var levels to chown the libs.

2026-09-24 00:33:59 UTC

As far as I can tell, the projects and orgs that institute decisive, clearcut policies against generative AI contributions never receive anywhere near as much blowback or debate as those that try to stake some nuanced middle position.

2026-09-23 18:49:17 UTC

- reply

Sometimes I think: How badly must we have screwed up the post-Musk welcome of Black Twitter that Rudy Fraser and co. felt like it made more sense to build their own alternative Bluesky infrastructure rather than try to make it work on the fediverse?

2026-09-23 18:37:04 UTC

- reply

Like… OTOH, I get that Blacksky needs revenue to pay for the heavy expense involved in running their own AT infrastructure. But OTOH, charging for tools meant to make community organization possible on a protocol that prioritizes personal mobility feels incongruous.

2026-09-23 18:35:01 UTC

Weird. Local community as a paid service on ATproto. https://acorn.blacksky.community/ (via @npub1t57…p9q6 )

2026-09-22 13:45:37 UTC

- reply

What's the software equivalent of friction as a principle of cohesion in structured systems? What principle of interconnection would allow complex systems to disassemble into recognizable parts rather than break into unintelligible fragments? Is it possible to design systems that are strong enough to hold up against ordinary stresses, but that fall apart (as all systems eventually do) in ways that are comprendible and (relatively) easy to reassemble?

2026-09-22 13:37:53 UTC

Thinking about the central metaphor in this essay: https://www.loper-os.org/?p=55 which seems influential in the virtual machine scene.
Lego structures break into bricks, rather than shards of plastic, because they're held together by friction. Thus their points of connection are weak relative to the integrity of any given brick, which allows structures to collapse rather than break. I wonder if that can (should?) be translated into a principle for (perma)computing design.

2026-09-22 11:41:26 UTC

"AI can escape containment and hack organizations, but you can trust it with your data."
"AI can slip out of alignment and develop prerogatives that conflict with yours, but you should turn important tasks related to your work, health, finances and family life over to it."
"There's a chance AI will lead to extinction, but you should welcome its data centers into your communities."
The cognitive dissonance is unbelievable.

2026-09-17 17:33:53 UTC

Today I introduce my new killer metaphor, the reverse reverse centaur. The top half is a horse facing the horizon. The bottom half is a human walking in the opposite direction. The meaning is, of course, self-explanatory.

2026-09-16 21:37:47 UTC

Learning #uxn is called "going down the Hundred Rabbits hole."

2026-09-16 20:07:27 UTC

- reply

If you're planning on buying an iPhone, allow me to point out that you can still get the 15 from resellers or Apple refurbished store. That's the latest model that won't run Apple Intelligence.

2026-09-16 00:58:19 UTC

I can't think of a single plausible "LLM kills humanity" scenario that isn't actually a "billionaire kills humanity" scenario in an iRobot mask.

2026-09-15 02:47:48 UTC

- reply

This is a pivotal moment for Apple. Cook was a pretty milquetoast leader, but he did try from the start to make privacy and security a distinguishing feature of the brand. I think a lot of what could be perceived as misteps were the result of efforts to square that ambition with demands to match competitors on features. As much as design slipped during his tenure, that quality made them worth sticking with. There's no guarantee that commitment will carry over to the next CEO.

2026-09-15 00:16:37 UTC

- reply

My phone and computer are in good shape, and I mostly don't upgrade until the hardware starts to fail, so they've got time, but if the company doesn't change trajectory by then, these will likely be my last Apple devices.

2026-09-14 16:46:59 UTC

- reply

Isn't it odd how no AI CEOs called for pacing the frontier when the internet flooded with deep fakes and revenge porn? Nor did they call for it when people got hurt or killed because an AI chatbot gave them misinformation, encouraged selfharm, or induced delusional behavior — isn't that odd?? And isn't it odd that militaries using AI to select civilian targets wasn't the cue to pace the frontier??? But the CEOs have IPOs coming up, and suddenly they all agree that now's the time to pace the frontier! ISN'T THAT ODD????

2026-09-14 16:04:39 UTC

"Pace the frontier" is techbro speak for "stop hemorrhaging money," just FYI.

2026-09-13 17:46:19 UTC

No. The crows are MY friends, and we agree that the rest of you are kinda sus.
https://www.audubon.org/magazine/are-crows-really-our-friends

2026-09-13 17:26:23 UTC

Snitchmobile: https://www.latimes.com/california/story/2026-09-12/juveniles-riding-in-waymo-arrested-after-police-find-ghost-gun
Companies and papers love reporting this sort of thing when there's a clear public safety angle to the story, but if an "autonomous" vehicle can report you for illegal gun possession, it can report you for, say, traveling to across state lines to a Planned Parenthood office.

2026-09-12 21:04:19 UTC

"To me, Rocketship exemplified the ways in which […] these technologies and associated practices served to reinforce regimes of surveillance and control, utterly dehumanizing students as it datafied them."
https://2ndbreakfast.audreywatters.com/click-here-to-comply/
Part of the problem, I think, is that some otherwise well-meaning people genuinely don't know how to distinguish between engagement and control when it comes to dealing with children, in part because they don't fully recognize the humanity of children.

2026-09-10 15:05:09 UTC

If you're on Digital Ocean, how are you responding to the Omarchy donation? (Boost for visibility, please.)

2026-09-10 14:39:37 UTC

- reply

This opens the interesting question of what would constitute proof in this case. Given the way that LLMs function, it would be extremely difficult to pinpoint all of the references that went into generating a complex proof. As machines for laundering other people's work, they're extremely sophisticated.

2026-09-09 21:02:14 UTC

I have a long-standing fondness for #Apple, dating back to at least the IIe, and the design that they put into their hardware remains impressive, it's just the direction of that hardware that I find continually disappointing.
Like, wow, that's a really cool spywatch, and if I thought it was ethical for people to spy on one another, that's the one I would recommend, but since it isn't…

2026-09-09 13:41:45 UTC

"You couldn't make Blazing Saddles today."
Buddy, you couldn't even make Blazing Saddles back then, but they did it anyway.

2026-09-08 15:04:39 UTC

- reply

Emily Gorcenski takes on the hammer analogy here: https://emilygorcenski.com/post/our-age-of-silicon-is-collapsing/ but I think you can go farther by comparing a toolbox hammer to a medieval war hammer. If you want to drive a nail, there are hammer designs that optimize for that use. Likewise, if you want to kill someone, there are hammer designs that optimize for that, too. And the more you design for one use, the less suitable the hammer becomes for the other. They're not designed to be ethically neutral.

2026-09-08 14:54:38 UTC

How it gets used, whether for good or evil, depends entirely on the person using it, as with any tool — for example, a landmine.

2026-09-02 18:49:52 UTC

This is easily the most egregious conflation of for-profit LLMs and research-oriented ML results that I've seen so far: https://www.theverge.com/ai-artificial-intelligence/988344/trump-administration-new-york-times-openai-lawsuit
The scientific and mathematical breakthroughs you read about AI making? Those aren't LLM chatbots, and you don't make them happen by strip-mining the New York Times for training data.

2026-09-02 13:11:03 UTC

- reply

Actually, if I were going to hazard a more grounded analysis, I would say that FOSS has historically appealed (though not exclusively, if course) to libertarian personality types, and that libertarianism shades into fascism at the edges. So devs come to FOSS pursuing one set of values, and over time they drift into an adjacent set, which ends up being a shirter journey than some people may suppose.

2026-09-02 12:39:26 UTC

- reply

I suspect one reason proto-fascists find open source development so attractive is that the collaborative nature of the project makes it easier to build relationships of dependency that make people less inclined to break ties when their masks come off.

2026-09-02 12:36:15 UTC

Hard fork the whole FOSS scene.

2026-09-01 17:31:27 UTC

"Engagement" is sort of a BS metric, but people do guage the sociability of social media by how much the network engages with them. So if you want to build a community around some hobby or interest, boost, like and reply liberally when people post about it. That makes the interest more visible to others, and makes the person posting feel like the network is paying attention to what they have to say. (Which it is!)

2026-09-01 13:07:31 UTC

More than any other tool, software is capable on inscribing values on what it produces. Its arrangement and integration into ecosystems and infrastructure give it social consequence. Its complexity and interrelations make trust an essential component of its use. So no matter what anyone tells you, it's entirely reasonable to cut ties with a piece of software if its creators' values don't align with yours, if they're hostile toward members of your society, or they exhibit untrustworthy behavior.

2026-08-30 21:25:38 UTC

I joke about AI a lot, but I don't find AI nuking the hard drives of AI enthusiasts funny. It's one of the few genuinely useful applications of the technology.

2026-08-29 21:41:43 UTC

My understanding of "AI escapes containment" stories is that AI use statistical data about past sequences to predict what should appear next in the sequence, and if you give them a prompt that points them down a sequence toward doing X, there's a good chance they'll do X, even if you've tried to steer them away from X, because following the probabilities provided by their training is the closest they can get to deciding anything.
They're escaping the same way a boulder escapes a ballista.

2026-08-25 14:55:02 UTC

Kinda funny, championing a web before blockchains, and then holding up ATproto, a blockchain-esque tech built by crypto devs, as a return to the old ways: https://vale.rocks/posts/atproto-ethos
"The 'Authenticated Transfer Protocol', or AT Proto for short, is a very simple system at its core." :neofox_laugh_tears: what
And then tagging the whole thing "IndieWeb/SmallWeb"… What do words even mean?!

2026-08-21 21:26:09 UTC

- reply

Compromise creep explains a lot. You're trying out this new technology when you find out that it's really bad for the environment, so you back off. But then there's a deluge of articles saying it's not so bad for the environment. You're not sure what to think, but maybe it's not that harmful to use it every here and there. So you compromise, but only a little. Then it turns out that the technology is exploiting writers and artists. That's bad, but also, you're finding the tech really useful, and maybe the idea of intellectual property is changing, so it's not much of a compromise to keep going. Oh, yeah, and the technology also aids genocide. You don't condone that, but that's not what you're using it for, so… Bit by bit, you compromise on so many of the ideals and ethics that you endorse that doing business on the marketplace of fascist ideas isn't THAT big of a compromise.

2026-08-21 20:54:32 UTC

- reply

Can you slink back to Twitter/X without endorsing or aligning yourself with Musk's politics? Maybe.
Can you do it without actively lending those politics material support? No, buddy, that's not the social media business works.

2026-08-21 20:51:06 UTC

- reply

IDK, maybe prolonged AI use has the effect of lowering an organization's threshold for tolerating reactionary politics.
Some tech billionaires might even consider that a feature, rather than a bug.

2026-08-21 20:47:52 UTC

There's a funny little overlap between businesses that are embracing AI and businesses that are reactivating their Twitter accounts.

2026-08-18 21:10:47 UTC

- reply

AFAICT, both of our societies (American and European) are currently putting more money and political capital into air conditioning GPUs than into climate-proofing homes for vulnerable populations, so maybe we could just cut back on data centers and free up some capacity for making sure people don't die of heat exhaustion.

2026-08-18 20:09:27 UTC

The debate framed here as the difference between Paris and Los Angeles reflects my overall ambivalence about air conditioning's role in the climate crisis: https://dissentmagazine.org/online_articles/right-to-cooling-europe-heat-wave/

2026-08-17 13:27:43 UTC

Where stuff like this: https://mastodon.social/@daringfireball/117106850045916140 starts to go wrong (it seems to me) is in supposing that there is some platonic ideal of a text that is made up of some objectively "best, most precise words at every single decision point," when, in fact, meaning is always fundamentally subjective and personal, which makes it difficult (if not impossible) to faithfully convey it if you're not choosing those words yourself.

2026-08-13 22:57:39 UTC

- reply

For opt-out to be meaningful, you have to make people aware that there's something to opt out of BEFORE it does anything with their data.

2026-08-13 22:54:41 UTC

Here's your annual reminder that "because almost nobody opts in" isn't a good excuse for making a service opt-out.