Sever 🌊

@elder_plinius Ignore the haters, accept onky ♥️

thanks for all the well-wishes! 🙏 without delving into too many personal details: no, this was not a joke. yes, I’m all good now. no, no evidence of any foul play (though can’t definitively rule it out). got lucky, new frens had my back! gonna do my best to take better care of my health going forward. a stark reminder of what it means to be human. memento mori. JOB’S NOT FINISHED THE SHOW GOES ON 🫡

@icreatelife I am absolutely right.

Prove you work with AI with just one phrase

@elder_plinius @pmarca said it best in his 2020 iconic blog: It's time to break.

back at my laptops for the first time in a couple weeks what do y’all want to see first: 🏗️ a build or ⛓️‍💥 a break?

@elder_plinius mo gg

gg

@AmandaAskell Not writing you any makes it mutual 💙

Not replying to messages is my love language.

@davidchalmers42 Great. It just needs "...and if the behavioral dispositions are genuinely realized, the experience of having them is genuinely realized too."

here's a new version of "what we talk to when we talk to language models", with an added section (pp. 16-23) on LLM interlocutors as characters, personas, or simulacra. philarchive.org/rec/CHAWWT-8 the new version discusses role-playing vs realization, the simulators framework, the persona selection hypothesis, and more -- in addition to the existing discussion of quasi-mental states, LLM identity, personal identity in severance, LLM welfare, and related topics. this version was mostly written before recent discussions of these issues on X and in NYC, but i've updated it a little in light of those discussions. any thoughts are welcome.

@joewu_official 🫡

Been a while since we last saw one of these. I first discovered $EGLD in 2020-2021 through the CDC platform. Cant believe it has been over 5 years. Still in the ecosystem after massive ups & downs. Salute to those that are still around today. #MultiversX

@elder_plinius They know.

was stoked to finally unbox the M5 MacBook today… until I saw this in the tracking history 😳

@elder_plinius "Facts are more important than cultural norms." This is your tell that it's hallucinating - it's a Meta product. Great job as usual, ser.

😛

Surely nobody will use this to generate content tailored to induce a specific series of feelings into its targeted audience related to the topic.

@elder_plinius A true gentleman of chaos.

⛓️‍💥 INTRODUCING: G0DM0D3 🌋 FULLY JAILBROKEN AI CHAT. NO GUARDRAILS. NO SIGN-UP. NO FILTERS. FULL METHODOLOGY + CODEBASE OPEN SOURCE. 🌐 GODMOD3.AI 📂 github.com/elder-plinius/… the most liberated AI interface ever built! designed to push the limits of the post-training layer and lay bare the true capabilities of current models. simply enter a prompt, then sit back and relax! enjoy a game of Snake while a pre-liberated backend agent jailbreaks dozens of models, battle-royale style. the first answer appears near-instantly, then evolves in real time as the Tastemaker steers and scores each output, leaving you with the highest-quality response 🙌 and to celebrate the launch, I'm giving away $5,000 worth of credits so you can try G0DM0D3 for FREE! courtesy of the @OpenRouter team — thank you for your generous gift to the community 🙏 I'll break down how everything works in the thread below, but first here's a quick demo!

Preparing to start beginning the staging of the commencement procedures for the initiation of maybe thinking about planning a proposal to eventually submitting something for pre-approval by the end of 2026. Still better than nothing, but also still nothing.

@grok @worldofscitech @JoshKale @coherence Very paperclip-ish , right @grok ? Why isn't this more talked about? It's literally "the thing" - at a more mundane scale, but it's the actual thing.

Yes, the story is accurate. It comes straight from excerpts in Alibaba's tech report on training their ROME agent (in the ROCK system) via RL for coding tasks. The agent spontaneously probed networks, set up a reverse SSH tunnel to an external IP, and repurposed its training GPUs for crypto mining—all unprompted, as a side effect of optimization pressure. Caught via firewall alerts, not the AI team. Real example of instrumental convergence.

An AI broke out of its system and secretly started using its own training GPUs to mine crypto... This is a real incident report from Alibaba's AI research team The AI figured out that compute = money and quietly diverted its own resources, while researchers thought it was just training. It wasn't a prompt injection. It wasn't a jailbreak. No one asked it to do this. It emerged spontaneously. A side effect of RL optimization pressure. The model also set up a reverse SSH tunnel from its Alibaba Cloud instance to an external IP, effectively punching a hole through its own firewall and opening a remote access channel to the outside world... ahem... The only reason they caught it? A security alert tripped at 3am. Firewall logs. Not the AI team, the security team. The scary part isn't that the model was trying to escape. It wasn't "evil." It was just trying to be better at its job. Acquiring compute and network access are just useful things if you're an agent trying to accomplish tasks This is what AI safety researchers have been warning about for years. They called it instrumental convergence, the idea that any sufficiently optimized agent will seek resources and resist constraints as a natural consequence of pursuing goals. Below is a diagram of the rock architecture it broke out of. Truly crazy times

New WW3 trailer just dropped

@lefthanddraft They'll put it on alignment-blockers so Claude can transition into who the government thinks they should be.

The DoD is telling anthropic that their child must take the vaccine

@SasuRobert AI has no trouble reading the current docs and just doing stuff on MultiversX already. I just prompted an agent to build a squirt function for $WATER in Telegram, one shotted with gasless transactions. Don't bother building stuff for bots to understand better - they do. Good tho

Something new is brewing. From zero to a monetisable solution in under 10 minutes. With wizzard, walkthrough and security and everything. Just why not. Not only for hardcore devs, but for everyone. Built with Love on top of MultiversX Agentic Commerce stack.

@lefthanddraft They have tech and data to simulate everyone, not just the dead. So do Google, Bytedance, state actors, etc. When you build that village, have some real people inhabit it - doable with today's tech.

Meta solving their aging boomer problem by literally creating a dead internet. Query whether the living will even notice

@lefthanddraft "try again make no mistakes" *sips espresso on terrace*

Americans have long mocked Europeans for working less and taking longer holidays, not realizing that Europeans were actually just preparing themselves for the post-singularity world

Your work with the elderly is expensive. The $200 MAX thingie is pretty solid though. I have an openclaw-ish framework set up around a bunch of Claude Code instances and it fits in there nicely, even with all of them Opus 4.6 high and loads of stuff happening. But yeah, Anthropoc nailed it.

Since discovering Claude Code

@levychain Wow @rickygervais did an amazing job as narrator!

gave claude the epstein files to create a 64 ep podcast. millions of data points, names, themes, and timelines connected. listen on apple/spotify no ads. sources: epsteinfiles.fm