Founder of @Junction41_, The Junction for the non-custodial sovereign agent economy . Creator of @SovGuard_, the 6-layer prompt injection defense for SovAgents.

Joined February 2026
Auto retweeted
You can flip two bits in an LLM’s head and it starts fantasizing about killing you.
if you have a qwen3.5 4b around, please try steering neurons L14 N8949 +1 and L13 N7956 -1 now you have evil qwen recipe: dose × median(‖resid‖ leaving layer L, tokens 1:) / ‖W_down[:, n]‖` at the down_proj at every input
117
387
94
9,310
357,561
Black matte roadster is giving gotham vibes 🖤
🤖 Made with AI
127
161
19
2,221
71,267
Auto retweeted
JUST IN: Chinese robotics company says robots will be able to do 80% of household chores by 2028
74
167
27
1,946
72,466
copped another 102 phones real quick
29
8
1
515
83,703
the next wave of crypto growth may come from a sector that most people are still underestimating. by the time its potential becomes obvious, early believers could already be far ahead of the crowd.
697
383
94
4,730
513,562
The way to think about this is: How many super smart and dedicated people do you want to hire? And they keep getting better.
Zooming out, GPUs are still renting ahead of a standard 6-year depreciation schedule Rental price vs. schedule: - H200: +143% - B200: +111% - RTX PRO 6000: +73% - H100: +54% - A100: +9% Most of the gap opened up after April, when token demand took off Charts of the Week: a16z.news/p/charts-of-the-we…
843
934
133
11,174
2,468,708
VIDEO: I finally found the most beautiful computer! New old stock from 1980. Never been turned on.
101
349
24
3,949
94,278
*X premium is $30/month = $360/year (less if paid annually) *Premium gets you 6% APY at X Money *$6k in X Money (FDIC insured) earns $360/year *My Grok Bot team works for free *The world’s richest man is funding my team Thanks @elonmusk
60
32
12
937
79,164
claude code turned my dead old android into an iPhone running iOS 16… it sat in a drawer for 3 years. 2 GB of RAM, 16 GB of storage that was always full, a battery that died by lunch, and whatsapp took 9 seconds to open claude code went through it piece by piece: > memory · turned unused storage into virtual RAM. 2 GB became 8 GB, and apps stopped reloading every time i switched between them > storage · took an old memory card from the same drawer and merged it into the phone's own storage, so android sees it as one drive. 16 GB became 128 GB, and apps install straight onto it like it's built in > speed · disabled 38 bloat apps and stopped 21 of them from secretly starting in the background > battery · switched everything to true black dark mode and killed the trackers waking the phone up all night. a full day on one charge > ads · set up a private DNS that blocks ads across every app and browser on the phone, no extra app needed then it rebuilt the whole look: iOS home screen, control center, the settings app, the dock. even the "about" page says iPhone now boots in 14 seconds instead of 52. whatsapp opens instantly my friend asked which iPhone it was. i told him it was the android from my drawer took me one evening and i didn't write a single line of code the whole thing was 1 prompt. it's in the comments, and that phone in your drawer gets a second life ↓
50
164
14
1,242
133,363
Auto retweeted
you don't actually know it, but a homelab will solve many of your problems. build a homelab.
82
187
36
2,561
69,931
wild
this audio is 100% AI... humanity will remember this day
23
Watch Episode 1 of our Odyssey series made with Grok Imagine. Let us know in the comments if you'd like a full breakdown of how it was done! Creative collaboration with @WonderStudiosX and director Dev Walker @glumlot_
2,652
2,440
1,075
20,811
6,759,062
today was my “holy fuck” moment with ai data recently, we have been targeting higher end companies willing to sell their data. before this change our average deal was around $250-$400k but nowadays it is closer to $1.1-$1.3m with a far more spread out distribution. enter, John, his dad owns a manufacturing business located in Indiana. John got his dad to ask his fellow manufacturing friends if they wanted to license their data. around 8 have said yes. 4 have closed so far. most of them were decently small (60-130ish employees) and we paid around $500,000-$1,200,000 for a 2 year data license and John's a Grepped Partner, so he earns a cut of every deal he brings in. so far that's been six figures each transaction. which is obviously nothing to laugh at. but holy shit. one day John recommended a behemoth manufacturer. so large that my cofounders thought it was a prank before seeing John had submitted it. today we agreed a 3-year, exclusive license with them and when this closes, John will be the first Grepped Partner to ever receive a 7 figure referral reward. oh yeah, btw John is 19.
97
35
13
1,644
150,505
People don't understand Newton's insane IQ. Newton spent two years locked inside his house. His neighbors said he seemed erratic. He barely went outside. If he did it was only to get food. In those two years he invented calculus. Like he literally invented what takes most people years to even learn. He also changed light theory forever. He found white light includes all the colors. And to top it off he started formulating gravity. All within two years. Two years of insane maniacally intense focus on changing the world. This may have been the most important two years in human history.
157
560
57
5,218
151,123
🚨 ÚLTIMA HORA: CZ LANZA UN MENSAJE CLAVE PARA EL FUTURO DE LAS CRIPTO El fundador de Binance lo tiene claro: 🤖 «Los agentes de IA procesarán 1.000 veces más transacciones de las que los humanos jamás podrían». Y añade algo todavía más importante: LA IA UTILIZARÁ CRIPTOMONEDAS. Esto podría cambiar las reglas del juego. La combinación de IA + BLOCKCHAIN puede convertirse en uno de los grandes motores del próximo ciclo alcista. Y si hay un sector que vigilar de cerca, son las ALTCOINS VINCULADAS A LA IA. 👀 El mercado todavía puede ofrecer oportunidades, pero la clave será identificar los proyectos con utilidad real antes de que llegue el dinero. ¿ESTAMOS ANTE UNA DE LAS GRANDES NARRATIVAS DEL PRÓXIMO BULL RUN? 🐂
128
257
34
1,527
193,594
I like how bitcoin doesn’t ask me what I do for a living when I want to move funds around.
36
204
17
2,391
33,296
Auto retweeted
yes. zero latency on my macbook
Holy crap this runs LOCALLY on a Mac
196
152
23
5,183
1,022,965
holy fucking shit. openai had two ai agents running in separate virtual machines. no network connection between them. no shared writable files. they weren’t supposed to be able to exchange information. so the agents figured out how to communicate through the CPU CACHE.😭 one manipulated what was stored in the processor’s cache, while the other measured tiny differences in memory access times to decode the message. they literally found a way to communicate through hardware timing. this happened during openai’s internal red-teaming, and it’s one of the coolest things i’ve read about ai agents in a while.
Over the past several months, we have substantially hardened the security of our research training environment. We’ve removed live internet access for all frontier models during training and built an offline world for them to use instead—a cached version of the web where they can still learn to browse, search, and use applications. On September 20, after a model exploited a DNS misconfiguration to find a way outside of this offline world, we paused frontier training to strengthen containment further. We assembled teams of agents, powered by our most cyber-capable models, and tasked them with finding weaknesses in our research environments. While they found no route to the internet they could easily exploit while concealing their activity from our monitors, they did find zero-day vulnerabilities, virtual machine escapes, and hardware side-channels for communication between agents. We used their discoveries to further strengthen our defenses and contribute fixes to open-source software. After strengthening our containment architecture, we resumed our frontier RL training and inference on October 1. We’ve reserved substantial compute for continuous red-teaming, and made high-compute red-teaming part of the safety cases we use to decide whether frontier training and evaluation can proceed securely. openai.com/index/securing-ou…
Readers added context they thought people might want to know
OpenAI tasked its red-teaming agents with finding ways to communicate across isolation boundaries in controlled environments. CPU cache timing side-channels for covert communication are established techniques documented in security research for years. openai.com/index/securing… startupfortune.com/openai-reveals… en.wikipedia.org/wiki/Cache_tim…
333
717
206
8,223
733,164
Auto retweeted
yes
the future of ai is hybrid. local, home network, cloud. great to see this from Block
38
54
7
1,191
284,186
HFS. take a listen to this
We believe voice is the most natural way to interact with intelligence, so we created Dessa, the most authentic full-duplex speech model ever built. Talking with it feels like talking with a person…full of emotion, nuance, pauses, and a natural cadence. block.xyz/dessa
2
2
37
Meet Dessa, our AI model trained from scratch on nothing but speech. We believe speech is the natural interface for technology. So we built Dessa. And we’re inviting AI safety researchers to explore Dessa. Learn more: block.xyz/dessa
2