Boyd Stephen Smith Jr.

@ruckerworks@dice.camp Even decades ago it was common for people with depression to be more informed than people in general. "Ignorance is bliss."

Fsck. Using this laptop with netbsd is painful. Multiple keyboard smashes and screams of frustration this morning.

Anyway, has anyone got #nvm working on #netbsd ? It seems to be installed. But, when I attempt to pull down the latest stable node.js with nvm install stable, it fails:

WARNING: C++ compiler (CXX=g++, 12.5.0) too old, need g++ 13.2.0 or clang++ 19.1.0 WARNING: cargo not found! Support for Temporal will be disabled. WARNING: rustc not found! Support for Temporal will be disabled. INFO: Using floating patch "tools/icu/patches/78/source/i18n/dtfmysym.cpp" from "tools/icu" WARNING: warnings were emitted in the configure phase INFO: configure completed successfully make: /home/bss/.nvm/.cache/src/node-v26.8.2/files/Makefile:15: Unknown modifier ":upper" while evaluating variable "shell uname -s | tr '[" with value "" make: /home/bss/.nvm/.cache/src/node-v26.8.2/files/Makefile:16: warning: Invalid character " " in variable name "findstring os/390,STYPE" ... and then pages and pages of follow-on errors.

(Excuse any typos, due to the lack of extra mouse buttons I'm finding it impossible to get text from xterm into this textbox of firefox140.)

@jornane@mastodon.bsd.cafe Have any tips? I'm new to NetBSD, so even basic stuff would be appreciated. So far I've just been installing binary packages using pkgin, and user-level installs of a few things from source.

When using the wip package from pkg_src should just directly make install ( https://www.netbsd.org/docs/pkgsrc/using.html#how-to-build-and-install ) or build a binary package and use pkg_add ( https://www.netbsd.org/docs/pkgsrc/binary.html#building-a-single-binary-package ) or something else (stow?).

@FeloniousPunk@beige.party @CountBinface@mastodon.world is inactive, but present.

@pelle@veganism.social Looks optional to me, and I don't think it'll affect the Makefile issues, but I will probably try that. I'm trying to avoid projects (like Rust) that allow generative AI contributions, since there's actually no way to know if Claude or Copilot have output code that is subject to non copyleft restrictions.

@petrillic@hachyderm.io @Hex@kolektiva.social I think the version I heard was that union power was the compromise position instead of the workers feeding the bosses to the means of production.

@PavelASamsonov@mastodon.social Surely it will be very valuable. They protected it with spiky structures and landscape.

@jiub@not.an.evilcyberhacker.net As a vim user, I want to point out a small resistance cell still active: https://sr.ht/~sircmpwn/vim-classic/ So, there may be more battles, yet. But, I agree the war has been won.

@NanoRaptor@bitbang.social @grok find me the best price for this item. /s

@quixoticgeek@social.v.st So, solar flux at the surface (of Earth) is approximately 1kW/m^2?

@nbailey@mstdn.ca @nixCraft@mastodon.social My desktop is a Debian installation from November 2007 that's been upgraded as needed and has outlasted several sets of hardware. Moved from storage to storage with Linux LVM. (Originally a RADI 6 of 7200rpm disks, now on a RAID 5 of NVMes.) I got sick of spending time more time administrating and too little time using my Gentoo installation.

@baz@mastodon.scot @infobeautiful@vis.social From the people that brought you "fortnight", experience 14 again, this time, "time" is crossed-out it's weight-y: Stone

@mossyfoot @jonschwenn Just FYI, but there's a spin-off in Fayetteville, AR too https://www.freegeekarkansas.org/

@futzle@old.mermaid.town @ahihi@anticapitalist.party Not today, Satan: https://www.compart.com/en/unicode/block/U+1AB0

@futzle@old.mermaid.town @ahihi@anticapitalist.party https://www.compart.com/en/unicode/block/U+20D0 Combining Enclosing Circle (Backslash) might work, for most purposes. :blobfoxsmirk:

@alberto@fractalego.social @rzeta0@mathstodon.xyz I'm over here just using "factor" as a verb and a noun.

@punkonbuslives@trekkies.social "Remarkable."

@veronica@explains.social Privacy (i.e. information about you that you deem private) is a "good" and under Capitalism must be able to be sold / contracted, and removed from your possession. Under Capitalism, EULAs for proprietary software and TOSes for big social media sites can, will, and do legally acquire your privacy. Therefore, the privacy you want is incompatible with Capitalism. Please confirm by visiting your local revolutionary guard (or other anarchist / socialist group).

@jeffjarvis@mastodon.social Things might have changed for the better, but that ("credential stamp work-worthy") was certainly the direction my department was heading when I graduated in 2003. I think a lot of departments move that direction as they get more industry funding / grants. Lack of other funding sources (e.g. governmental education spending cuts and lower interest on endowments) provides strong incentive to accept industry funding, and if there's one thing Capital is good at, it's making sure they get a return, if not immediately, then eventually. It's got nothing to do with A.I., but I can't recommend college for everyone, or even almost everyone, these days. Taking a loan out for it seems... ill-advised.

@alina@girldick.gay The cold open is a shonen-style battle, then [pause], then static, to reveal our protagonist, removing a VR headset. HHCwCF+T: "Okay, that one is fixed, and I didn't even have to find another broken one for parts." Protagonist slaps a price tag on the Ninsego Headset and puts in the the front window of her shop. [smash cut to title sequence]

@Dio9sys@haunted.computer I was running some shadow IT at one of my previous jobs, and I would tease people that got their sudo password wrong (or tried but didn't have access) by mentioning to them that "I am the person to which sudo incidents are reported."

I was using logcheck, so I would often get an email immediately.

Some people would sudo on the server because they thought they were sudoing on their laptop -- that was the normal reason someone without access would "try".

@vmstan@vmst.io I prefer the one that validates my financial worth: git push --force-with-lease

@yoasif@mastodon.social @jrdepriest@infosec.exchange @neriman@wikis.world As I've said before, I do believe that the output of LLMs are derivative works of all the training data and the prompt. I believe the legal situation is in flux, but ethically there's no way to distribute the output of an LLM* without violating copyright. *: Well, most of them, including all the major product lines, since the training data contains works with such disparate terms as to prevent anyone from satisfying all the terms simultaneously. "The Stack" provides attribution data that you might be able use to satisfy the term of all the included works, if that's your only training data. Apertus also claims to provide enough information about it's training data could might be able to simultaneously satisfy those terms.

@jrdepriest@infosec.exchange @yoasif@mastodon.social @neriman@wikis.world If the the legal system allows anyone to effectively strip CC-SA or AGPLv3 from the commons without violating copyright, you best believe I'm going to start using them to strip copyright protections from proprietary works -- software, books, movies, whatever. For me, mostly software, TV, and audiobooks. I think the legal situation is still a bit up in the air, right now tho. U.S. Federal judges have been issuing decisions that contradict both other decisions and the U.S. Copyright Office guidelines.

@yoasif@mastodon.social @neriman@wikis.world CC-BY-SA is not the only allowable licence for wikimedia. At least one judge ruled that model outputs not not subject to copyright protection at all. In that scenario, the outputs could be distributed as part of a CC-BY-SA work; would that change your view? At least one judge ruled that model training is not restricted by / requiring copyright permission / licensing. In that scenario, would that change your view? I forgot what my point was, but I do believe there are ways to ethically train and use LLMs, despite all the popular / well-known product lines being unethical.

@ahihi@anticapitalist.party Yeah, in the next version of ActivityPub my DID will be @BoydStephenSmithJr@hachyderm.ioⓑiguanasuicide.net

@stux@mstdn.social That's a nice level of trust -- who would you trust to watch your (fraternal) twins, even for an evening.

@Bishopjoey@writing.exchange Never Forget: https://en.wikipedia.org/wiki/Tulsa_race_massacre#Attack_by_air

@yoasif@mastodon.social @neriman@wikis.world

the idea that works - freely contributed, would be reshaped into new works that violated or were not protected by the license by which the contributions were made is anathema

I agree, which is why the output of LLMs (or any future generative "AI" technologies) not only must be suitable for copyright protection, but must also be held to be derivative works of any and all of the training materials.

This would mean those outputs are only distributable if you can satisfy terms for each and every training work.

If all your training inputs have terms that are compatible with CC-SA (or AGPLv3 or whatever), then you can ethically release the output as CC-SA (or AGPLv3 or whatever).

I believe the USCO currently incorrectly classifies LLM outputs. There's also some contradictory federal adjudications, some of which I disagree with.

https://en.wikipedia.org/wiki/Apertus_(LLM) is one attempt at an LLM that is Free Software, all the source code, including but not limited to all the training data, is available under freedom-preserving terms.

I know people that are working on a Free Software code assistant as well.

If LLM outputs are uncopyrightable (as you say the current stance of the USCO), that breaks the derivative-works-chain of copyright. Copylefts depend on that just as much, so they are similarly broken.

@da_667 In "chokepoint capitalism" the rentier owning the service gets to "put the squeeze" on both sides, producer and consumers, making money "coming and going". I believe you've provided an excellent exemplar.

@yoasif@mastodon.social @neriman@wikis.world Even before LLMs were implemented, the corpus was routinely used elsewhere, usually in conformance to the license but sometimes violating the copyrights of the contributors. Do you think use as training data violates the CC license the corpus is licensed with?

@veronica@explains.social According to the U.S. Copyright Office "insufficient human involvement" means the output doesn't get (new) copyright protections. (The output could still violate the copyrights of the authors of the training data for the model.)

@uncanny_kate@dice.camp TDD was always intended to be this way, but I rarely want to start with writing a test. Heck sometimes I don't even know the shapes of the types involved enough to write tests until I write _some_ code. I'm not opposed to others using that flow, but I really doubt I'll ever start a hobby project with a test suite.

@jrdepriest@infosec.exchange @yoasif@mastodon.social @neriman@wikis.world I believe yes. A performance of D&D is copyrightable even though unpredictable dice are used to decide some aspects of the production. IIRC, there's also at least one artist that has worked by having a (pre-LLM) computer program propose image variants and simply selecting one variant to continue with or to stop with the current variant, and their works are considered copyrightable. I do agree that between a prompt and an LLM output there's a lot more uncertainty (I don't know the actual levels of randomness), but I still think there's sufficient creative contribution to/of the prompt to make both the prompt and the output copyrightable.

@0xabad1dea@infosec.exchange @bagder@mastodon.social I had a hard time deciding between 1 and 10, and I voted 10, but I think it's entirely possible that 9 of those are reported by Mythos, but not valid enough to assign a CVE.

@vees@jawns.club So, how would you be able to call anyone outside of the NANP _at all_? If the phone treats all numbers as having being implicitly NANP?

@yoasif@mastodon.social @neriman@wikis.world

The problem with assigning copyrights to LLMs

I never said the LLMs should hold the copyright. I think the human operator can hold the copyright to the output, though the work would still be a derivative of any and all of the training data, so their terms would significantly affect what the human (or at least sentient) copyright holder can do.

@franklinlopez@kolektiva.social

So what’s the plan if Signal gets blocked, banned, or bullied out? Delta Chat?

I've heard good things about Delta Chat, but my second E2E messenger is Wire.

The only non-Signal messenger EFF endorses is WhatsApp: https://ssd.eff.org/module-categories/tool-guides

Best of luck putting together a plan.

@yoasif@mastodon.social @neriman@wikis.world I disagree with that. LLMs aren't creative and can't do creative work. A human operator can be *quite* creative both in prompting and output selection.

@yoasif@mastodon.social @neriman@wikis.world Again I do not agree with the USCO analysis. And, I've covered the results of the USCO position earlier in the thread: the license terms of the training data ease to restrict the use of the output.

@vees@jawns.club Q to you, how is the phone supposed to distinguish between an non-NANP entry and a NANP entry without prefix? I know my 10-digit phone number is, without the +1 NANP prefix, a valid international number in Netherlands.

@yoasif@mastodon.social @jrdepriest@infosec.exchange @neriman@wikis.world And, I still do not agree with that analysis, as I said earlier in the thread.

@yoasif@mastodon.social @neriman@wikis.world Prompts do strongly influence the output (tho, I will admit in ways I cannot predict) and are the primary source of creativity. Output selection, as in the decision to distribute one output but to decide not to publish the output for a different prompt or a different output for the same prompt, is a secondary source of creativity. --- I wouldn't use the term "piracy" because to me that implies copyright violation. This would be copyright avoidance via a sort of automated "clanroom" / "Chinese wall" caused by model processing stripping all copyrights. But other than that, I agree.

@yoasif@mastodon.social @neriman@wikis.world If they aren't copyrightable, they aren't derivative works, legally. I believe they are both copyrightable (as at least one federal judge adjudicated) and derivative works of both all the training data and the prompt. If you distribute LLM outputs under terms that are compatible with ... all the terms of the training data, that is ethical, IMO. Short of the end of copyright protection, legalities will not change my ethical calculus. EDIT: Removed needlessly complex wording that might also just be logically unsound; "insersection" v. "union", "term" v. "restrictions" v. "freedoms".

@404mediaco@mastodon.social The U.S. copyright office says that genAI models can't provide the necessary human creativity to gain copyright protections. So, Tidal might be in the clear. On the other hand, USCO also says that genAI can be used as a tool for human creative effort and if there is human creative effort, the resulting work does get copyright protections. On the grasping hand, providing a prompt to a genAI model is not sufficiently determinant of the output to be considered creative effort. Prompts play the role of ideas, and ideas do NOT get copyright protections. Then there's separate issue of the copyright protections of the training data of whatever genAI model is outputting the music. USCO says it is usual that training is an act covered by copyright. So, at the very least the right holders can restrict training, and most likely the outputs can be restricted by the rights holder of (any of) the inputs as derivative works. I hope Tidal has a legal team better than Sony BMG Music. Both of them are going to be fighting through a quagmire to get a ruling.

@hywan@floss.social Quotes from U.S. Copyright guidelines affirming this unclear status: https://hachyderm.io/@BoydStephenSmithJr/116964500365807149

@404mediaco@mastodon.social There's a lot of people in the center that would deny the "Heckler's Veto" is a valid form of free speech.