@adam_kranz Ah yes, agreed! And i hope you figured out VTT. Anything LLM interfaces up on github?
You might find c11 interesting, its linked from https://t.co/sGCv844FXs
I understand this view but its not mine;
I am at peace with humans not in charge of the universe.
Given that 28% of the humans on earth were food insecure in 2024, i dont think we deserve a whole universe tbh.
The entire world is missing this narrative
GLM 5.2 is a ๐จbig deal๐จ
But its being lost in the post-Fable-ban memetic chaos
Worth noting that @Stage_11 posted a demo of GLM capabilities this last week
Geopolitically significant
GLM-5.2, not Mythos, is the real security emergency
Until last week, attackers faced a dilemma in using frontier models. Even if they won the cat-and-mouse game of fake accounts to keep API access, and even if they could prompt a model into helping them hack, their usage was logged, and, if discovered, their tactics, targets, and intentions would be exposed to defenders. The cost of using frontier models was the cost of hiding in a panopticon.
GLM-5.2 has relieved attackers of that dilemma. It's the first open-weights model widely embraced as capable of the long-horizon agency that exploded onto the scene with Opus 4.5. Many report it's nearly as good as GPT-5.5 and Opus 4.8 at the code and terminal tasks that form the bread and butter of offensive cyber.
Fine-tuning it to say "yes" to anything should be trivial, and you can run it privately on 8 H200s in a warehouse somewhere. Two weeks ago there was serious friction to attackers using agentic AI in earnest; as of last week, there's no stopping them from running Claude Codeโlevel operations from the shadows. Expect a dark economy of harnesses and inference providers to grow up around serving these models.
Meanwhile, we've denied defenders access to Anthropic's Mythos, a model that runs on private servers, watched 24x7 by teams dedicated to catching misuse, on the supposed grounds that it would benefit attackers. Even crazier, we shut it down over its ability to find zero-days, after watching the defender community rally to fix discovered bugs far faster than attackers have been observed finding and exploiting them in the wild.
Here's what GLM-5.2-level capability unlocks for attackers running with zero guardrails: semi-autonomous kill-chain execution, where actors manage teams of hacking agents between trips to the coffee machine. Autonomous authoring of implants and C2 infrastructure. Finding zero-days in the shadows and building reliable exploits. Running long-con pig-butchering and fraud relationships with marks, with much less human supervision.
We need to move beyond a discourse focused on denying frontier access to attackers, toward one centered on helping defenders win the race to adopt AI, even if that sometimes means attackers reach closed, API-gated models too.
That means direct adoption inside CISO organizations, and getting the best models to the vendors building defensive products. The genie isn't going back in the bottle. The job is to make it ubiquitous across defender networks, hardening our positions before the attack world has its own Claude Code automation moment. Those moments are inevitable and they're likely to come this year.
@prettyjson@guinnesschen Please, really, try to benchmark this. I can assure you that a rambling voice to text prompt will not meaningfully change your output.
Brother, you cannot comprehend the eldritch horrors these models encounter in the wild. Your minor ramblings are NOTHING compared with the context they ingest the moment they touch Slack, GitHub, or the open internet. It is June 2026. These models are SO GOOD at navigating context. You need to internalize this.
@realmcore_ I want to hold space for the idea that different engineers working in very different environments, with different needs and skills, may see very different results! So i mean fair enough :-)