foxyoreos (they/them, commissions open)6d ago
None of the fuckers promoting this shit understand security, they are so one-track-minded that they will give an AI chatbot the ability to send password reset emails, and then they want you to take them seriously when they go "oooh we're gonna scan your repository for vulnerabilities" #AIBullshit

RE: https://bsky.app/profile/did:plc:vk7rduhvom3rq6dyluce5wzf/post/3mnaku3pf422w
Show threadfoxyoreos (they/them, commissions open)6d ago
I need you to understand that the *same people* who do this shit are trying to lecture you about the future of software. And so much of the AI "debate" revolves around pro-AI sides being able to flip a switch and imagine these people are suddenly magically respectable and serious. #AIBullshit
Show threadfoxyoreos (they/them, commissions open)6d ago
The people who wire up an AI chatbot that lets you hijack accounts are the SAME people who are trying to sell you on agentic workflows, and all of it falls apart if you don't close your eyes and clap your hands and manifest in your imagination that they're suddenly good at security. #AIBullshit
Show threadfoxyoreos (they/them, commissions open)6d ago
I've been called naive for pointing this out before - I've had pro-AI accounts tell me that *of course* people are working on prompt injection, *of course* it's gonna get better, but when you press them, there's nothing behind that except magical thinking and hopes and dreams. #AIBullshit
Show threadfoxyoreos (they/them, commissions open)
My introduction to AI skepticism was red-teaming the original ChatGPT 3 and early ChatGPT 4, years and years ago, and realizing, "holy shit, this can't be used in a practical way in any serious application" and in all those years.. NOTHING about the security has changed. #AIBullshit
Jun 1 at 8:53pmWeb
Show threadfoxyoreos (they/them, commissions open)6d ago
"AI is suddenly useful for programming-" Every single agent is vulnerable to unpatchable security flaws. There is no new state of the art to make a safe agentic workflow. Nothing has changed since GPT-3. None of it works unless you stick your head in the sand and pretend you're safe. #AIBullshit
Show threadfoxyoreos (they/them, commissions open)6d ago
Anti-AI critters get criticized for "denying reality" a lot. But I am watching the tech industry look at an unpatchable security flaw that renders this tech unusable for most secure applications, and they're just kicking their little feet and screaming "NO, stop SAYING that" #AIBullshit