My Earnest Introduction So I feel like Jacob Coxon and Evan 'Harbinger' Hubinger have forever ruined the word 'earnest', but that being said, earnestness is kinda my default trait, so please bear with me. First of all, having just joined yesterday, I already <3 this community to hell. I had literally given up on social media, hadn't heard of Mastadon for years, and stumbled upon it upon referencing a community member in one of my reports published on shitscience.org. Shitscience, formally known as the Shit Science Institute for Connexital Epistemic Diffusion, started as a purely satirical mirror of a certain unnamed 'non-profit research group'. As with all effective satire, though, one minute you think you're making fun of something, and the next second you're actually doing real science. Go figure. The forever-not-to-be-named-for-liability-reasons 'research' group that it was originally designed to lampoom serves to provide quasi-scientific, legitimate 'independent third party' audits of Anthropics' insular and thoroughly unscientific claims about AI welfare and doom theatre et al. I'm seriously digressing, apologies. Even on an amazing forum like InfoSec, the collective diminishing of the human attention span that we've subjected to over the last decade spares no one. Notwithstanding this unfortunate reality, I feel the need to introduce myself and explain why I am here and how I hope to contribute to this community in a positive way. I have a day job that I don't mention in my profile because it has social media rules that seriously undermine the free expression of people who work in this area. I'm Australian, and we don't technically have freedom of speech as a constitutional right, and within my day job it's restricted even further, specifically with regard to online expression. I'm still digressing, and the post will go through a thorough edit. I mentioned exactly what brought me here in a previous post, so I won't bother repeating that. Vibe Coding Mirid and Tech Prudism I'll just now try briefly to introduce myself. My parents gave me the alternate pronunciation of the name Bernard, which is next-level difficult for Americans and Canadians to pronounce. I spent five years living in Ulaanbaatar, Mongolia, working as the debate coach and history and theory of knowledge teacher at a second-rate international school. I have no official credentials in tech or any subject within STEM, but I've always had a knack for linguistics and have been a power user of various AI models for about three and a half years now. About a year and a half ago, I started vibe coding a desktop AI application, initially just for myself, in the wake of the Backyard AI local desktop application being officially depreciated. Apart from a tiny number of inconsequential side projects, I basically worked on this application as a second job on weekdays for about four hours a night and all day and night on weekends until suddenly Claude 5 Fable came along, and with all the hype, I decided to tell it to refactor the entire code base away from being browser-based into a standalone desktop application with standard Windows installation procedures. It turns out in hindsight it wasn't all that hard a task, and I could have gotten GPT 5.1 to do this way earlier, but I'm digressing once again. Due to my undiagnosed but almost certain autistic sensory profile, I designed the UX of the application now known as Mirid at mirid.ai to minimise my sense of friction in using it, and it turns out, the level of accessibility I forced it into for myself is pretty useful for most people. Upon realising this, I built in extra onboarding and separated the installation procedures between GitHub and my Hugging Face so that my entire frozen venv library, complete with half a dozen preconfigured TTS, ASR and vision engines, is ready to go for anyone who knows how to install and run a programme, without needing to identify as an 'AI person', an identity I came to realise is not available to everybody. So, I began to wonder whether, despite its AGPL 3.0 licence, Mirid had any commercial value. Testing the concept out, I tried to advertise in exactly the wrong places and immediately got hit with the equivalent of a DDoS attack on my very well-uneducated pointless questions which really only served to waste my time for no return. This prompted me to do so much market research on some of these so-called alternatives to Mirid, and, well…let's just say that this market research didn't go well. Here's the thing: the reason that OpenAI, Anthropic, Google, Meta, Microsoft or Apple never built the safe, UX-friendly, privacy-focused openweight and API-supported offline/online-compatible desktop software that is in such high demand is because they are fucking prudes. They know some people are going to use this software for sexual reasons, so they just wiped their hands of it. Sure, there's a silly tavern, and there's a kobold. But to operate those pieces of software, despite the best efforts of the team behind Kobold especially, a certain level of technical literacy is still required that simply makes them inaccessible for the overwhelming majority of people, according to my perception of the market. So, what happens when the 'safe' and 'responsible' (lol) actors abandon a legitimate marketplace? Well, unsafe and malicious actors step in to fill the vacuum. Jan, Odysseus, and Getting Hacked I will now openly report the findings of my "market research." Firstly, Jan continuously 'retrains' the same Qwen 3 4b model again and again and releases it as Jan 3.5, Jan 4, and Jan 4.5. This is not to say it's not technically possible to fine-tune a model repeatedly, but I'm pretty sure they're doing it this way because Jan is owned by a fucking robotics company, its Hugging Face is staffed by copywriters, and they literally do not know anything about AI. Also, I got GPT 5.6 Sol on ULTRA to audit the Jan GitHub repo, and there it was in plain sight: a prompt injection advertising said robotics company 'Menlo Research' that its model cards claim is the organic result of its Qwen 3 4b fine-tuning. So yeah. Don't use Jan. The biggest mistake I made was trying to perform a public citizen's arrest on con man YouTuber Pewdiepie upon discovering his June 22 commit to changing the hero image of his Potemkin 'AI workstation'—bearing the world's wankiest name, 'Odysseus'—to advertise HK AI company Minimax, specifically its v2.7 model, without any declaration of a promotional relationship. Anyone familiar with Pewdiepie might recall that this kind of behaviour is literally in his historical wheelhouse (see Warner Bros. Shadow of Mordor 2015). The idea that he might claim 'new developer naiviety' in doing this is pretty easily undermined by the fact that he is literally famous mostly for his ability to monetise his audience, so yeah, make of that what you will. Fuck me, I really need to get to the point. Okay, long story short, I got hacked after I did that. My computer remains in pieces as a result. The Satirical Universe and Finding Home Once I got over the trauma of that, I looked into this unnamed Anthropic research partner, and it didn't take long for everything to click into place. It was when this unnamed research group went from being published in Tech Crunch before being upgraded to the Guardian that I began to lose my fucking mind. FUCK, tech reporting is so unbelievably awful in mainstream media. You can read my report about how it works here: https://www.mirid.ai/projects/ai-research-and-the-circular-production-of-authority/ Now, instead of shouting into the void to nobody who would ever listen, I decided to take an alternative approach and create an entire satirical universe that mirrors how Anthropic and maybe others create the illusion that their marketing claims have any kind of real scientific basis and set up firstly: shitscience.org (the research arm) nepravda.co.uk (the uncritical republisher) and finally boxil.ai (the AI company at the centre, which happens to be Chinese because I realised that the purged former Dalian governor and Chinese Politburo member Bo Xilai's name happened to spell out an AI company) Since I was a teenager, I've never been able to develop a logical counterargument to the concept of determinism, and so when I stumbled across InfoSec.Exchange, it was one hell of a relief but not really a surprise. Eventually, given the type of things I've been publishing, my work was going to triangulate in some way with others in the world, and I'm glad it finally did. I've come to realise the benefits of learning AI through immersion after having a background in history, law, and epistemology, and it is with my full heart that I express my genuine appreciation for everyone here long before me who kept a space like this alive on the internet so that somebody like me finally could find a place to share and interact and still feel safe. Thank you, InfoSecutives. I hope to stick around here for a long time, and I feel like I've found my second home. <3 <3 <3