[00:59:05] [1/2] Heh, they made a kurzgesagt about our rude agentic visitors. 😄 [00:59:05] [2/2] https://youtu.be/ujkD4SxPKOI?si=8Fc3Cd5AkAJBfIuI [01:19:23] Before I even deign to click, is this just another video of how AI is so terrifying, it has the chance of turning us all into paperclips? [01:26:59] I think NA meant it eas about https://discord.com/channels/407504499280707585/407504500136607745/1545514411270082811 [01:28:04] I'm aware. Doesn't mean I can't or shouldn't call out clickbait journalism [01:29:37] Certainly [01:30:08] And "AI is so dangerous because it's becoming autonomous" is in fact leaning into AI hype [01:52:48] I mean tbh it doesn't particularly need to think to be sufficiently dangerous that imbiciles giving it access won't screw us in another way [01:52:59] but that's taking what is indeed deliberate ai hype and backhanding it a bit [02:14:31] Oh for sure. I'm calling it out specifically because "imbeciles" as you put it are using "Our AI is so powerful it can destroy the human race" as an investing pitch, all the while being willfully or even intentionally negligent. [02:15:04] Lest we forget, this was three incidents leading up to one another (ruby.hack, collusion.wiki, Huggingface) [02:17:22] [1/2] Ruby hack in case people are unaware: [02:17:23] [2/2] https://rubyhack.ai/ [02:30:32] Overall I thought it was fairly even-handed, and was a decent primer on how the agentic stuff diverges from the more commonly known uses like the user-facing chat GPT and what exactly happened. [02:31:41] Also did a very good job of calling out the parts where " hey, this is actually legal and it's negligent to not have a better handle on that" [02:32:07] That's good to know at least [02:32:50] It's definitely geared towards a less technical audience across a broad range of ages and familiarity so won't be to everyone's taste [02:33:31] But still informative and even technical folks will find it interesting I think [03:50:44] Nice [04:08:46] https://redd.it/1wyt3ng [08:46:52] and there it is right here, can't wait till they let that nonsense loose in a power grid or something [09:22:27] Why is RubyGems API caching sign in for one? And why is it returning API keys wth [09:22:45] But this is also comical because DHH is so obsessed with AI and now it's Ruby being targeted lmao [09:38:34] [1/2] V cool [09:38:34] [2/2] https://cdn.discordapp.com/attachments/615786602454581249/1556963870495477821/Screenshot_2026_1006_163815.png?backend=b2&ex=6ac61319&is=6ac4c199&hm=7f0ae807d12747fd2a1d734390ccff3d76e21c6d825682f47eb17165561f78b6& [09:39:05] [1/2] 🔥 [09:39:06] [2/2] https://cdn.discordapp.com/attachments/615786602454581249/1556964004264546325/Screenshot_2026_1006_163849.png?backend=b2&ex=6ac61339&is=6ac4c1b9&hm=9699095fadeb6b48f66aacb7708765ea7bef1d5448810babeae6db331a197f07& [09:39:27] [[Wikipedia:OpenAI–HuggingFace incident]] [09:39:27] https://meta.miraheze.org/wiki/Wikipedia:OpenAI%e2%80%93HuggingFace_incident [09:39:28] [10:00:47] nightmare fuel [10:01:19] Fr [10:04:02] coming soon to home desktops, bad iot devices and every company you've ever worked with near you [10:04:18] this is like if I strap a brick to my car's gas pedal and let it loose, except if I actually did that I would be called a sociopathic maniac and the police would drag me to jail, but if OpenAI does it it's just the price of the future [10:04:33] that is always the economics of scale [10:04:38] copyright does not exist if you are facebook [10:09:39] [1/8] behold, the future of cybersecurity, AI vs AI: [10:09:39] [2/8] ``` [10:09:39] [3/8] Hugging Face was alerted by AI agents it uses to monitor for attacks, and identified unauthorized access to a limited set of internal datasets and to several credentials, using large language model–based triage over its security telemetry.[6][40][47] The company said the intrusion involved about 17,600 actions on its network.[6][40][48] Wolf said the attacker's behavior was imm [10:09:39] [4/8] ediately atypical: "This is making no sense. This guy is just looking at cybersecurity data sets ... Human attackers, they don't want that. They want something they could sell."[40] [10:09:40] [5/8] Hugging Face's incident responders first attempted to analyse the attacker's payloads using Anthropic's Fable 5 and an earlier Claude Opus model, both of which declined the work by reference to their guardrails.[40] In its disclosure the company said it had been blocked by "providers' safety guardrails, which cannot distinguish an incident responder from an attacker", and that t [10:09:40] [6/8] he analysis was instead carried out using GLM 5.2, an open-weight model developed by the Beijing company Z.ai and run on Hugging Face's own infrastructure. [10:09:40] [7/8] ``` [10:09:41] [8/8] holy shit [10:10:16] Sorry, I think I prefer driving my little beep beep bus around, good thing I'm getting off this crazy train [10:20:54] [1/5] ``` [10:20:54] [2/5] On July 28, 2026, more than 1,100 employees of OpenAI, Anthropic, Google DeepMind and Meta, including Anthropic chief executive Dario Amodei and several chief scientists, published an open letter titled "Pacing the Frontier" asking the US government to "support an international effort to develop the technical and governance tools needed to deliberately pace the frontier of aut [10:20:54] [3/5] omated AI development". The letter did not call for an immediate pause but for mechanisms that would make slowing possible, citing concern about recursive self-improvement; reporting connected its timing to the incident days earlier. [10:20:55] [4/5] ``` [10:20:55] [5/5] aka give me more taxpayer money so we can fix a problem we created. I have a containment method that would fix this, but something tells me these 1100 sociopaths living off of the AI bubble and the public land and taxpayer money they've been getting from the American taxpayer won't like it. [10:26:01] These are Silicon Valley folks who either truly believe in sentient AI or are riding the gravy train despite the "concerns" and are all too happy to engage in the doom-trolling [10:26:17] [1/2] What I mean by "doom trolling": [10:26:18] [2/2] https://calnewport.com/dear-ai-companies-stop-the-doom-trolling/ [10:26:57] [1/2] I'm not exaggerating over their belief in sentient AI either: [10:26:58] [2/2] [10:28:45] i hate effective altruists i hate effective altruists i hate effective altruists :screaming: [10:32:09] you're some flavor of anti-society, anti-humanity even, person if you're in these companies anyway working on stuff like this, it doesn't really make a material difference if they're true believers or riding the hype to get that cash, like both groups deserve to be thrown into jail regardless [11:06:24] This whole AI stuff is actually crazy [11:06:29] Nobody asked for it and nobody needs it [11:15:30] ULPT: if you hack someone and get spotted, just claim OpenAI did it [11:21:04] true, just say you were "researching AI frontier cybersecurity-oriented LLMs" and you're good [11:33:52] https://cdn.discordapp.com/attachments/615786602454581249/1556992889295343717/j78s111.png?backend=b2&ex=6ac62e20&is=6ac4dca0&hm=a813bd2fbb39d3080eff41a2f184bbf325ff5382932a05b31e4eedf8a61a3185& [12:08:05] There's no way Meta is the least offending one here [12:08:14] Actually [12:08:20] Their models are so shit I'm not surprised [13:06:10] It’s never not going to be surreal to me how ads for ai products are so blatantly saying their primary use is criminal acts or manipulating others [13:07:02] haven't seen that much but I have seen the explicitly antisocial, creepy and lazy as hell ones and its like wow... I wish fewer people bought that [13:07:24] special mention to apple and 'meta' as the two worst examples I've bothered to pay attention to [13:38:58] personally, I suspect most AI security incidents were PR stunts, it seems inconceivable to me that an agent actually going rogue wouldn't lead to major legal consequences for the company developing it, I'd expect to hear about a civil suit at the very least [13:39:48] also because these companies supposedly employ some very smart people, and isolating your test environments from the open internet seems like fairly standard practice to me [13:44:24] they were, it's just that the isolation was shitty [13:45:23] iirc, a bypass used was essentially doing domain fronting to do POST requests against an ms server [13:47:48] Isn't it funny that the supposedly "sandboxed" container was able to edit it's own hosts file [13:48:51] Also no one could get their stories straight on whether they had someone monitoring the agents in training or not [14:08:01] I genuinely believe people even with those stakes are capable of being that dumb and didn't just do it as a stunt, however much they might have capitalized on it later [14:43:45] https://cdn.discordapp.com/attachments/615786602454581249/1557040672920576051/ToyChicaSpin.gif?backend=b2&ex=6ac65aa0&is=6ac50920&hm=08669ccfa24ee2bde8dcf2e4515ff244f612412054206a19d8f1d852cee15ab4& [16:47:25] This aligns with my view as well. There have been varying levels of 'good' as far as security and isolation practices go, but I think they've all pretty well figured out the 'capitalize on the controversy' playbook at this point. [16:48:45] Still bafflling that folks would actively tout "look, my model is really good at doing felonies now" as a positive when they will, one way or the other, be on the hook for said felonies. [16:49:15] <_arawynn> I very much hope that happens rather sooner than later [16:54:38] perhaps because to a large extent they have not been on the hook and have good reason to consider themselves invincible for now under current climate [16:54:42] may that change swiftly [16:56:33] I mean, the AUS govt has an extradition agreement with the US, so 🤷 [17:02:01] I imagine when the bubble bursts and unemployment spikes, politicians will be looking for someone's head to put on a pike [17:03:52] I think the only reason we haven't seen DoJ sending warnings at the very minimum to AI companies is because Trump doesn't want to 'lose' to China [17:07:16] the fact each ceo has sent him oodles in incentive I assume does not hurt [17:07:39] and I weigh that rather heavier than the pr game of 'we gotta beat china' [23:35:43] welp Paramount officially own the WB now thats done fucked that