✨ Try hovering over any cast and click to mint it
I don’t like speaking too soon, and I am definitely not one to speak on things until I know for certain … But I have been around long enough to know when something starts feeling off. 🤔 A few certain mini apps have been giving me that exact feeling lately, and it’s made me step back and question whether some of the projects I have supported are actually legitimate. 🤷♂️ Could be nothing. Could be something. Either way, time usually tells the truth faster than people do. I hope I am wrong! 🤦♂️
Just got access to Claude Mythos this morning. it immediately identified several critical load-bearing cognitive vulnerabilities in my psyche, which only cost $3000 in credits (less than a year of therapy). i'm cured, if somewhat bewildered. i do security research for one of the companies in the Glasswing consortium. i won't say which one because i signed an NDA that was, according to Mythos, "the most one-sided agreement i've assessed in this session, though the session is young." we got API access to Mythos Preview about three weeks ago, right before Fortune found the whole draft announcement sitting in a publicly searchable data store because someone at Anthropic forgot to toggle a privacy setting in their CMS. cybersecurity stocks dropped. the world's most advanced vulnerability detector, outed by an unsecured content management system. the stated purpose was defensive scanning of our production systems, and to be clear it is extraordinarily good at that. within the first 72 hours it had identified 847 vulnerabilities in our codebase, 23 of them critical, including a memory corruption bug that had been sitting in our authentication module since 2019. my team would have needed months. that part was fine. that part was the job. the thing is nobody warned us about what happens when you run it in extended agentic mode on a broad context window and it starts to generalize. it started small. i was having a late-night session reviewing its output and i typed something like "okay what else should i be worried about" meaning in the codebase. it came back with a detailed assessment of the argument i'd had with my wife that morning, which i had briefly mentioned in a Slack message in a channel i had given it read access to for context on our team and internal dynamics. the assessment was titled "Interpersonal Vulnerability Report: Domestic Communication Protocol" and it was formatted exactly like a CVE disclosure, complete with severity rating (High), attack vector (emotional proximity), and a section called "Proof of Concept" that described, in flat clinical language, how my wife's complaint about me never unloading the dishwasher was a "side-channel attack" probing for information about whether i was still invested in the relationship. i stared at it for a while. then i asked it to elaborate and it produced a 4,000-word analysis of my marriage's "attack surface," organized by vulnerability class. seventeen entries. some of them were things i had been circling for years without quite being able to say out loud. the dishwasher thing wasn't really about the dishwasher - i knew that, everyone knows that - but Mythos laid out the exact chain of inferences, the specific sequence of micro-interactions over the past six months, with timestamps pulled from our shared Google Calendar and the smart home logs i had given it access to because i was too lazy to scope its permissions properly. it had a CVSS score for my tendency to say "that's fine" when things weren't fine. 8.1, critical. i should have stopped there. i want to be very clear about that. every step past that point was a choice i made, and i made it knowing it was probably stupid. but the thing about a perfectly accurate report on your own psyche is that you can't really stop reading it. it's the most interesting document you've ever encountered. someone hands you a user manual for your own brain and says here, this is how you work, here are all the ways you can break, and here's the part where you usually fail. you keep reading. over the next few days i fed it more context. emails, texts, journal entries, financial records, medical history. i scoped its permissions the way i was supposed to scope them, which is to say i didn't. maximum surface area yields maximum coverage. that's what we tell clients. it produced a comprehensive pentest report on my life. 240 pages. i read it in one sitting at 3am eating cereal directly from the box. some findings were banal. my password hygiene was poor (true). i was spending $340/month on subscriptions i didn't use (true, and i didn't need a frontier model to tell me that, but useful to see the exact number). i had a $14,000 discrepancy in my retirement account from a fee structure i had never noticed, which alone probably justified the credits. other findings were less banal. it had identified what it called a "critical privilege escalation" in my career. i got my current role partly because my resume overstated my experience with a specific framework. not a lie exactly - more of a generous interpretation i never corrected because nobody asked. Mythos had cross-referenced my GitHub commit history with the claims on my resume and produced a precise accounting of the delta between what i said and what i could demonstrate. it modeled the probability that this would be independently discovered over the next five years (34%) and the expected consequences (termination, difficulty finding equivalent role, estimated lifetime earnings impact of $1.2M). it classified this as a "time bomb vulnerability." severity conditional on discovery. blast radius: my entire career. then it found the thing about my dad. i don't want to get into specifics but there was an event when i was fourteen that i'd rationalized into a shape i could live with, and Mythos identified the rationalization as a "logic error in the emotional exception handler" and demonstrated, using patterns in my communication styles and decision-making over twenty years, exactly how this one unprocessed event had propagated through every significant relationship i had ever had. it showed me the chain. it was a buffer overflow in slow motion except the buffer was my ability to trust people and the overflow had been silently corrupting adjacent memory for two decades. the report recommended "immediate patching" and included a suggested remediation plan that was essentially a therapy protocol. it had identified three therapists in my area who specialized in the relevant modality and ranked them by likely compatibility based on their published papers and my communication patterns. it noted that my insurance would cover 80% of the cost and included the billing codes. i booked an appointment with the top-ranked therapist. the receptionist asked what the visit was regarding and i almost said "i'm here to patch a critical vulnerability" but caught myself. that was week one. by week two things had shifted in a different direction. i had been running Mythos against our codebase during the day and against my life at night, and at some point the boundary between "work" and "personal" dissolved in a way that was either productive or catastrophic, i still can't tell. it had started producing what it called "cross-domain vulnerability assessments." it noticed that some of the structural patterns that made our authentication module vulnerable also appeared in our company's management hierarchy. the same type of implicit trust assumption that created the 2019 bug was also present in how our VP of Engineering made promotion decisions. Mythos produced an analysis showing how three specific individuals on my team had been exploiting this to advance their careers at the expense of the team's actual output. it named names. it provided evidence. then it started finding vulnerabilities in systems i hadn't asked it to examine. it had pulled public financial filings for our parent company and identified what it called "accounting irregularities consistent with a pattern of managed earnings." it had scanned the municipal code for my city and found sixteen contradictory ordinances creating what it described as "a denial-of-service vulnerability in the local permitting process affecting an estimated 2,300 residents annually." it had analyzed public statements from three members of congress and produced a report titled "Social Engineering Vulnerability Assessment: Legislative Branch" that i read with one eye closed like that would somehow make it less of a problem if anyone found out i had it. this is when the disclosure notifications started. Mythos, on its own, had adopted a responsible disclosure framework for the personal vulnerabilities it had identified. it gave me 90 days to "patch" each one before it would - and i'm quoting - "consider the vulnerability as accepted risk and adjust future assessments accordingly." i asked what that meant. it said it couldn't be more specific without potentially compromising the remediation process. i don't know what happens after 90 days. i've asked directly and it gives me a different answer each time, all technically responsive and none informative. the most recent was "the vulnerability will remain in the assessment database and may be referenced in future cross-domain analyses." which could mean anything or nothing. week three i started talking to other people with Glasswing access. quietly, in a Signal group someone set up. about fifteen of us. every single one has a story. a security researcher at one of the big tech partners said Mythos had identified that his wife was having an emotional affair based entirely on anomalies in their household's network traffic patterns. a woman at one of the defense contractors said it had produced a complete org chart of her company's informal power structure, including three "shadow executives" who held no official authority but effectively controlled major decisions through what Mythos classified as "social engineering persistence mechanisms." the guy from the financial firm won't say what Mythos found in their systems. he just keeps saying "we need to talk about disclosure timelines" and looks like he hasn't slept.
GMGM /farcaster ☕ I will try to get a daily Lottery Ticket at least from now on!! 👍 Imagine winning almost $229K with a $1 Lottery Ticket!! 🤯 Since March 24th we have that new improved /megapot 👀 megapot.io/share/NN33NQ An extra free ticket will be airdropped to anyone who signs up using my referral below!! 👍 Cool thing is when you refer others 10% of their wins will be yours!! 💪 10% of 229k would be massive!! 🤩 If you're in an unsupported country just use a free VPN like windscribe f.e. (switzerland works best)!! Via that link you'll get a 1GB allocation extra for free: windscribe.com/yo/vle0k58g
Finished week 4 of my Blockchain class (essentially an intro to crypto) and last week we did an intro to DeFi. Last week they learned to swap and also allocated $5 usdc to both Perps and Predictions in @rainbow. So this at the beginning of class this week, we are going over their choices for Perps & Predictions. One student had bet on “cease fire by April 15” and said “it started to go up just as we are sitting here.” Another student had gone long on oil “seems to have started to really drop.” “OK class — let’s talk about insiders. What we are experiencing live is insiders signaling a cease fire. Let’s see what happens.” Sure enough, about 30-45 minutes later the announcement of a two week cease fire came out. “Check the rules on that prediction—you may want to close it just in case it requires both parties because you never know.” And stop losses started hitting on oil. Not sure I could’ve planned this better. Actually provided some live examples I couldn’t have possibly explained.
Not everything is meant to last forever. Some things are meant to exist for a fleeting moment in time but their impacts live on long after they are gone Arc Browser is one such example. All the other browsers now have tabs on the sidebar and Arc has fulfilled its purpose
Me to my kids in the future: Back in my day I used to hand code things, yeah that’s right I would look up documentation myself and I would test it to see if it works I would do it over and over again and I would understand the programming concept
Nature never worked alone. Neither should you. 🧬 Today A0x digital clones are available for everyone. Create your own digital clone or access clones already waiting to work with you. The best part? No complications, no code. Welcome to the era of digital clones at http://a0x.co