Elaine Jackson-Pimentel 🦄🌈

@prahou@merveilles.town I unplugged the Alexa and said how dare you speak our holy tongue!!!

@SwiftOnSecurity@infosec.exchange That explains those weird OpenBSD users who act like Navy Seal Team Six is going to break down their door any second now!

@malwaretech@infosec.exchange That makes sense.

@shinspiegel@mastodon.social @ki@chaos.social @paco@infosec.exchange The LLM thing is a bit overblown. Contrary to popular beliefs, large language models do have practical size limits in terms of being able to perform inference. If we collected every single chat message, social media message, every keystroke of every user in the world, the language model would be too big. So instead of a library of all of the information in the world, you need the most popular and most relevant information. Both xAi and Perplexity both use retrieval augmented generation, a process where they collect information from their index and pass to the prompt, of social media posts rather than including them in the model.

@globcoco@mamot.fr @paco@infosec.exchange Do you think your bank is better at cyber security than Google? You give them the same information. The alternative is giving photos of your photo ID to random websites with no full-time security team. If the data is collected who has a objectively better cyber security team? Google or a random adult videos website. We do have an alternative to age verification and it's censorship. It could just be illegal to transmit material harmful to minors across state lines. No age verification because there's no material to age gate. Unless you're willing to actively advocate that there should be zero safe guards to prevent a small child or teenager from being exposed to the materials, for the sake of adults having easier access to the material. Governments are going to do something. We can either present them with options that avoid harm as much as possibl, or we can stay silent l, and let them decide and it'll probably mean everyone loses. You might have to provide a government ID to use the internet, and there's censorship.

@paco@infosec.exchange I agree we have to take things with a very careful approach. I don't think leaving content on the internet that could cause developmental harm and past generational trauma is a good idea to expose minors to. If age verification is just too risky then our next option is to conseuer the content. Removing the ease of access to harmful content such as pornography is an alternative we could consider. We could pass strict liability laws that can't be bypassed and include personal criminal liability to board members of data is breached. The status quo of clicking a yes im 18, would be like a gas station selling alcohol to teens and asking y'all are 21 right wink-wink! Some has to change and that's either safely verifying age over the internet and if that's not possible then removing it from the internet n I will close with the fact that to open a bank account online you have to provide your drivers licenses ID number and your social security number, if banks can do it securely so can the rest of the world.

@paco@infosec.exchange I don’t understand why operating system developers aren’t required to implement age verification. Apple, Microsoft, and Google could coordinate a standard that passes age group data to websites without revealing identities. It makes no sense for every app or site to handle this alone. Linux and Firefox users could use Google’s web service linked to their accounts. Problem solved. Everyone wins. Only these companies have the security scale to manage threats in real time.

@mmorschel@mastodon.social Many businesses and people want a centralized model as the it's on Elaine's NextCloud or Ashley's NextCooud makes it harder to find things quickly. It enables faster building but at the cost of vendor lock-in. I looked at Nextcloud only a very long time ago. Maybe I'll check it out again in the future. Search is a harder problem, DuckDuckGo has been the subject of controversy as they added AI features their users didn't want. This effectively meant DuckDuckGo was sharing every search query with an AI model which is yikes! People expect it from Google but even the private tools do stuff which is borderline malicious at times.

@mmorschel@mastodon.social I worry about a future where the internet becomes a silo. E.g. Europeans are not allowed to talk to the rest of the world. Or even algorithms are required to prioritize local and suppress international content. And this can also happen with restrictions on accessing US Tech. Everyone deserves the right to choose their tech tools. I want to see #foss earn serious adoption because it gives us an alternative to everything is now a subscription service. That can't happen if they ban GitHub and GitLab limiting international cooperation. And I agree only having USA Tech means that they now have leverage over the rest of the world. Someone could probably make a fortune by copying Gmail and Google Docs and not having AI features and charging a fee instead of data collection for advertising.

@malwaretech@infosec.exchange I hope to see you include recommendations that provide some potential guidance for working on these systems to defend against malware abuse.

@donhawkins@mastodon.social @nixCraft Courts interpret and apply laws passed by Congress. They don’t invent them. The fair use doctrine, codified in 17 U.S.C. § 107, has existed for decades. It isn’t some recent invention crafted to favor AI companies. Separately, courts have ruled that Anthropic’s use of pirated copies did not qualify as fair use. This resulted in a major settlement. Many creators, authors, and legal scholars strongly disagree with how fair use has been interpreted in recent AI-related cases. Appeals or future litigation could still shift the balance. That said, courts are simply doing their constitutional job: resolving disputes according to the laws already on the books. Dismissing judicial rulings just because the outcome feels unpopular or inconvenient undermines the rule of law itself. In the end, while tech giants can afford armies of lawyers to navigate, litigate, or simply outlast these rulings, it’s everyday creators, small authors, independent artists, and regular people like you and me who will feel the real squeeze. Our livelihoods and our rights to control and benefit from our own work are quietly eroded in ways the mega-corporations with deep pockets and endless resources never will.

@nixCraft@mastodon.social The claim paints AI (like ChatGPT) as a thief destroying creators, but recent U.S. court rulings (e.g., Bartz v. Anthropic and similar cases in 2025) have found that training large language models on copyrighted works is highly transformative and qualifies as fair use under copyright law, as long as outputs don't reproduce near-copies of originals. Courts called it "spectacularly" or "quintessentially" transformative—shifting purpose from consuming the work to learning patterns for new generation. This isn't "theft"; it's legally protected innovation in most analyzed cases (though piracy in acquisition can still be an issue). Critically evaluate AI outputs yourself—don't treat "ChatGPT said" as authoritative. It's a tool, not a source; responsible use means cross-checking, especially for facts, code, or creative work. Traffic drops to sites like Stack Overflow, forums, and wikis stem from efficiency: people now get quick answers via AI for common queries, bypassing searches. Stack Overflow's question volume and traffic have declined sharply since late 2022 (some reports cite 50-78% drops in usage/questions by 2025-2026), as developers turn to conversational AI. This mirrors Google's featured snippets and instant answers (pre-AI era), which pulled from Wikipedia and reduced clicks/traffic to sources—yet Wikipedia editors kept contributing, and infrastructure load eased. Viewers often still clicked through for depth. AI accelerates this trend but doesn't erase the value of human-curated knowledge for complex or novel problems. AI's impact on visual/creative fields is overstated here. Text dominates LLM use; AI video/image generation sees far lower everyday adoption, and music is niche—most people (myself included) stick to human-curated platforms like YouTube Music or Spotify playlists. Stats show AI tools help producers (e.g., generating elements), but full AI songs remain a small fraction; human creation drives charts and culture. Blogs still thrive, people read and share personal posts, and social media amplifies independent voices more than ever—no gatekeepers like publishers required. Human culture persists robustly. Social platforms buzz with original content, memes, art, discussions, and activism. AI handles rote language tasks but can't replicate human cognition, empathy, lived experience, or true innovation. Reasoning models mimic steps, yet falter on nuance, originality, or context without human grounding. For big problems, background knowledge and critical thinking remain essential—AI augments, not replaces. This isn't "destroying creators" or completing a cynical circle; it's disruption, like search engines or calculators before it. Innovation often shifts workflows—some lose, others gain—but human creativity endures and adapts. Blaming AI wholesale ignores nuance and evidence. Be a critical consumer: use tools wisely, support creators directly, and recognize evolution isn't erasure.

@ottaross@mastodon.social I think this is a step in the right direction. There is some demand for AI products that promise more privacy. Assuming that DuckDuckGo is running their own LLM or even paying OpenAI or similar for the API and assuming that OpenAI or whoever acts in good-faith and doesn't save chat contents and assuming DuckDuckGo doesn't share information about you it's a nice to have feature. This does however mean sharing every search query to an AI company which is concerning. Google already has your search query so adding an AI answer doesn't share your data with anyone, but for DuckDuckGo this is a transfer of personal information which is a biger issue. Maybe a better alternative would for it to be opt-in and you'll be prompted at the top of your first search to "Enable AI Features" or "No, don't ask me again (you can always change in settings."