Post by @cxiao@infosec.exchange

RE: https://mastodon.green/@gerrymcgovern/117172792450861828

there are no "ethical LLMs", regardless of local models, regardless of open weights, regardless of renewable power,
regardless of country or origin

you cannot have models this large and this "capable" without the accompanying mass abuse of human labour

Gerry McGovern

@gerrymcgovern@mastodon.green 22 hr. ago

‘In the end, you feel blank’: India’s female workers watching hours of abusive content to train AI
https://www.theguardian.com/global-development/2026/feb/05/in-the-end-you-feel-blank-indias-female-workers-watching-hours-of-abusive-content-to-train-ai

Invisible labour behind 'intelligent' machines
https://theloop.ecpr.eu/the-invisible-labour-behind-intelligent-machines/

Exploited Data Workers in Global South Are “the Secret Ingredient of AI”
https://www.democracynow.org/2026/8/27/data_workers

Inside Venezuela’s Hidden Workforce Training AI Algorithms for Pennies
https://www.democracynow.org/2026/8/27/ai_platform_extractivism

Inside the world of Kenya’s ‘shadow scholars’ paid to write essays for UK students
https://www.theguardian.com/education/2025/sep/14/kenya-shadow-scholars-paid-to-write-essays-for-uk-students

2 Replies

9 Boosts

William B Peckham

@wbpeckham@techhub.social 6 hr. ago

@cxiao Don't give them that much credit! Ethical LLMs are perfectly possible, they just chose a quick-and-dirty snatch and grab disregarding and ignoring all laws and rights because that was faster and cheaper. They are unethical crooks, not victims of immature technology. (Although their actions have made most of the rest of humanity victims of immature technology.) it would have taken longer, but they could absolutely have built their large language models based on only public domain and contributed text and verbiage. They could have built their image libraries out of only public domain and contributed images. (And creative commons-licensed images.) They could have, but they chose not to, take the time to curate the data and build an ethical database properly. They were in such a race for the money that they decided to cheat instead.

Vincent B. 🌱

@vincent@floss.social 21 min. ago

@wbpeckham @cxiao I’m now really wondering what to think about Apertus 1.5, because it genuinely seems like it checks those boxes (and is a lot less powerful than frontier models as a result), and I’m wondering if run locally on long-lasting hardware, on green-ish electrical grids, it could be a reasonable, actually ethical tool, worthy of trying to find positive uses for it (and that’s another problem).