@jedbrown@hachyderm.io

· 4mo ago

Excellent thread on what "LLM-assisted" testing actually does, with context from rsync replacing its test suite with a poorly vibe-ported Python version (as regressions roll in)

More generally, the primary product of LLMs is deception: the unfounded belief by the insufficiently-critical observer that the task has been performed. From vibe-coders to user communities, managers, and investors, there are many targets of the deception. Many are playing along for their own self-interested reasons (e.g, executives need investor confidence more than they need a product). And many feel hit when peers and communities see through it.

neuromatch.social

jonny (nonvenomous): "I think the modal situation here is that the peop…" - neurospace.live

75

4

43

1

Jed Brown @jedbrown@hachyderm.io

· 3w ago

RE: @jonny@neuromatch.social

17 GWh at US eGrid average of 350 gCO2e/kWh is 6 kilotons CO2e, plus 5.1 ktCO2e scope 3 (amortized supply chain) = 11 ktCO2e. That's like 2 weeks of emissions for an entire university with 50k people, including commuting, travel, etc., or the emissions of 16700 round-trip transatlantic flights.

US data centers are increasingly running on behind-the-meter gas with lower-efficiency simple-cycle gas turbines, which raises the estimate to 15 ktCO2e.

neuromatch.social

jonny (nonvenomous): "let's do some more back of the envelope estimates…" - neurospace.live

4

0

1

0

Jed Brown @jedbrown@hachyderm.io

· 1mo ago

Thanks to Deborah Lupton for organizing this panel "Resisting GenAI: How and Why", and to my co-panelists Dagmar Monett and Zachary Kaiser. I hope you enjoy and can make use of what we had to say.

https://www.youtube.com/watch?v=eOplEr_isKc

Resisting GenAI: How and Why

3

0

1

0

Jed Brown @jedbrown@hachyderm.io

· 3mo ago

Replying to

@faithisleaping@anarres.family And it rarely stops at personal use and standing behind the (mid, at best) results. It's always followed by forcing slop on others and demanding they welcome it, forced adoption by others, demands that nobody disapprove of their newfound psychological dependence, insistence that all critique is fake and calls for government to suppress "well-respected people" from criticism, etc. When the oligarchs and colonizers obtain a new weapon, wield it they shall.

6

0

1

0

Jed Brown @jedbrown@hachyderm.io

· 5mo ago

I have often said that LLM-based pull requests are a denial of service attack on mentoring in open-source communities. This post explains the rationale behind Zig's prohibition and offers much-needed reflection on the human surfaces of open source.

https://kristoff.it/blog/contributor-poker-and-ai/

kristoff.it

Contributor Poker and Zig's AI Ban

You should always change doors when playing Monty Hall

10

0

8

0

Jed Brown @jedbrown@hachyderm.io

· 6mo ago

RE: @olivia@scholar.social

This is how the "AI in government" people talk about using LLMs to generate text for regulations. They're proud of it. It is a betrayal of the public trust.

“We’re flooding the zone.”

"Google Gemini can do word salad."

https://www.govexec.com/technology/2026/01/government-ai-trump-administration-plans-write-regulations-using-artificial-intelligence/410946/
(linking a partner publication because the ProPublica Guild is on strike to limit executive demands about "AI")

scholar.social

9

0

6

0

Jed Brown @jedbrown@hachyderm.io

· 12mo ago

Replying to

@adventure_tense@mapstodon.space Yes. The article shows how they've been practicing good stewardship:

Austria isn't just replacing Microsoft software. Unlike typical public-sector and corporate migrations, Austria's military has heavily invested in LibreOffice development itself. The armed forces have been funding the creation of new features and improvements that are now included in public releases. These additions, ranging from improved slideshow editing to better handling of pivot tables, have been rolled into the latest version of LibreOffice.

@petersuber@fediscience.org

25

6

19

0

Jed Brown @jedbrown@hachyderm.io

· 3mo ago

RE: @GossiTheDog@cyberplace.social

This KPMG report contributed to the inevitability narratives that have been used to pressure academia. This is what businesses need, workforce development, etc. University administrators and business school faculty are exceptionally gullible marks.

It's notable that the report has been out for 8 months and this was only detected because an "AI" vendor (GPTZero) had an adversarial interest. So much for the discernment capabilities of business/finance/executives.

cyberplace.social

Kevin Beaumont: "KPMG issued a report citing all the transformatio…" - Cyberplace

3

1

0

Jed Brown @jedbrown@hachyderm.io

· 6mo ago

Replying to

@jenniferplusplus It's also important that to whatever extent this product actually works (I'm as skeptical as you are), it fundamentally preferences the attacker. The product has way too many false positives to run in CI, so the defender can only use it as part of an occasional audit. The attacker doesn't care about CI or development friction, and wins by finding one exploit in an entire stack, even if they have to wade through many false positives to find it.

5

1

0

1

Jed Brown @jedbrown@hachyderm.io

· 5mo ago

I installed new Challenge tires on my bike today. Please clap.

First time since the 1990s that I've needed a tire lever (and I worked in a bike shop for a year).

4

0

Jed Brown @jedbrown@hachyderm.io

· 5mo ago

Replying to

I am struck by the privilege and serendipity of a cold email recieving prompt and thoughtful personal replies from both study authors. This is utterly unsustainable, yet even that was not sufficient to overcome the anchoring effect of the epistemically-nihilistic affirmation machine.

3

0

Jed Brown @jedbrown@hachyderm.io

· 7mo ago

Replying to

@ricci Refusal warms my heart. We're trying to get shared governance recognized to produce an ethical use policy. Basic compliance with regent law. @SnoopJ @timnitGebru

4

0

1

0

Jed Brown @jedbrown@hachyderm.io

· 6mo ago

Is it good that a bunch of universities signed contracts saying they are responsible for everything students do with the product and that the university will indemnify OpenAI?

"Florida officials investigate ChatGPT, OpenAI over alleged role in FSU shooting"

https://www.nbcnews.com/news/us-news/florida-officials-investigate-chatgpt-openai-alleged-role-fsu-shooting-rcna267477

nbcnews.com

1

0

1

0

Jed Brown @jedbrown@hachyderm.io

· 6mo ago

Replying to

@jonny I love getting a section-by-section breakdown of all 13 sections of a (nonexistent screenshot of a) "contract", with direct quotes of the most concerning parts and pages of recommendation for negotiating revised contract terms.

1

0

Jed Brown @jedbrown@hachyderm.io

· 30mo ago

Replying to

@AlanSill@mast.hpc.social @johannes_lehmann@fediscience.org The infrastructure does not need to be expensive. https://joss.theoj.org/about#costs

joss.theoj.org

Journal of Open Source Software

Journal of Open Source Software (JOSS) is a developer friendly, open access journal for research software packages.

5

1

0

Jed Brown @jedbrown@hachyderm.io

· 19mo ago

Replying to

@hipsterelectron@circumstances.run @micr0@fuzzies.wtf @esm@wetdry.world Ah, but you see, it's only public domain after being laundered through "AI", which as a political/power consolidation project, means you have to pay a company for an approximate copy of the original. Making everything public domain would have some redeeming qualities not shared by the AI-laundering model.

1

0

Jed Brown @jedbrown@hachyderm.io

· 5mo ago

Replying to

@bkuhn@fedi.copyleft.org I had browsed the docket, but you are right that it is not for me to say whether motions are a delay, and plaintiffs also do not seem to be in a rush (e.g., joint motion to postpone deadlines). The point is that we don't know how such litigation will play out, especially in light of the volatility of public sentiment about this industry. Has anyone written an analysis of how their case pushes toward compulsory licensing? If LLM outputs routinely constitute derivative works, then it is impossible to comply with licenses (even permissive ones) without acknowledging all such training data and/or constant open-ended research quests as due diligence that each response does not infringe an unknown corpus. The companies don't want to disclose their corpus because their business relies on not acknowledging the derivative relation. @cwebber@social.coop @ossguy@fedi.copyleft.org @richardfontana@mastodon.social

0

Jed Brown @jedbrown@hachyderm.io

· 5mo ago

Replying to

@cwebber@social.coop I agree about the hazard. LLM outputs should be considered derivative of all their inputs unless established otherwise. LLMs manipulate expression, not ideas, and the propensity of verbatim reproduction (up to and including entire books) is evidence of that process. Note that the purpose of the "substantial similarity" test is as circumstantial evidence of process. I think the counterpoints are "mutually-assured destruction" and/or "yolo denial-of-service attack on copyright will win because power likes it". "AI" companies are still delaying cases from 2022 (like Doe v GitHub) because they want a jury who believes it is inevitable. Plaintiffs seek to win their cases, not to establish broad precedent. OpenAI has already lost (in German court) on copyright infringement of their outputs, arguing unsuccessfully that the infringement is the sole responsibility of their customers for prompting. The political reality of public sentiment is changing and collapse of the financial bubble will greatly alter the power held by "AI" companies. Meanwhile, I think the words of the DCO ought to mean something, even for those who are certain they are a smol bean. https://hachyderm.io/@jedbrown/114931171543347621 @bkuhn@fedi.copyleft.org @ossguy@fedi.copyleft.org @richardfontana@mastodon.social

hachyderm.io

Jed Brown: "What does the Signed-off-by tag mean? It is certi…" - Hachyderm.io

0

2

0