RE: https://dair-community.social/@timnitGebru/116659484433359308
the industry wants us to believe the “black box” is mystical rather than statistical, because mystery sells and helps them defer accountability.
This #TIME page is a reminder that “AI” is the equivalent of a pop-up ad. No one ever wanted them either.
RE: https://wandering.shop/@cstross/116648209688298468
The #AICon is off the charts
Very foolish. Very dangerous.
A Reddit post about the frustration of a developer who contributed #emacsclient support to #Gemini CLI, only for #Google to transition the tool to an enterprise-only, closed-source model shortly after.
[…] My takeaway is that in the future I will think very carefully before contributing again to non-FOSS projects.
https://www.reddit.com/r/emacs/comments/1tkn8sw/emacsclient_landed_in_gemini_cli_and_why_i_wont/
Yes. Also, the small problem of Google and AI use in genocide and war crimes, but not a priority I guess.
On how to unconvincingly weasel your way out of a big LLM 💩 …
[…] Mr. Rosenbaum’s book contains many quotes that are accurate, but the misattributed and invented quotes are scattered throughout.
[…] In his statement, Mr. Rosenbaum said that if the episode “serves as a warning about the risks of A.I.-assisted research and verification, that is why I wrote the book.”
https://www.nytimes.com/2026/05/19/business/media/future-of-truth-ai-quotes.html
#AI is a marketing term, not a technology. Preferred term should be the term #llm because that’s what it is. Also, I’m pretty sure this will be normalized soon enough and we will be told to just relax and accept this equivalent of ultra-processed food as acceptable compromise.
Brazilian Lawyers Fined R$84,000 for Prompt Injection in Court
… possibly first judicial sanction of attorneys who attempted to manipulate AI legal tools through adversarial input injection.
Source in Portuguese https://www.jota.info/trabalho/juiz-multa-em-r-84-mil-advogadas-por-prompt-injection-para-manipular-ia-usada-no-trt8
Microsoft Copilot Cowork 🙄 automates malicious prompt injections
This part especially:
[…] Scheduled tasks increase the risk surface for attacks like this significantly, as the user is not present to stop malicious workflows, and the prompt injections can take effect on a recurring basis.
https://www.promptarmor.com/resources/microsoft-copilot-cowork-exfiltrates-files
[...] data-driven business models of the traditional web (e.g., advertising, analytics) are being replicated in LLM ecosystems with limited oversight... All four platforms embed third-party tracking scripts in their web interfaces. Conversation URLs, page titles, and user identifiers are transmitted to ad networks — in several cases regardless of cookie consent.
#aicon #llm #ChatGPT #Claude #Grok #Perplexity #Meta #Google #TikTok
oatmeal
Sir Dr Rusty o the Isle 🖤💛❤️
Captain Jack Sparrow
