Robotics and Automation, often dreaming about the nature of nature.
Globetrotter
Joined September 2023
- Tweets9.8K
- Following826
- Followers501
- Likes78.8K
trap — e/acc 🏴☠️ retweeted
i’m giving one of you a fucking ROLEX
like + repost + follow me
one random winner picked 48 hours from this tweet. no purchase necessary
trap — e/acc 🏴☠️ retweeted
I didn’t upload for 23 days 😭
Retirement was insane.
I owe a Lamborghini to anyone who can guess what happens in the next YouTube video. I fucked up.
I’m officially back. New YouTube video this Saturday, October 24th. 👀
trap — e/acc 🏴☠️ retweeted
I feel like getting over engaged in the AI safety discourse is a massive psy-op that only assists whatever fuckwitery is going on to push up the value of certain companies.
When you boil it all down, what's really happening is:
1. Yes, AI is in many cases able to hack more persistently and more creatively than your average attacker.
2. Decades of poor cyber practices are now all coming to a head. Anyone in cybersecurity has been shouting about unpatched servers, flat networks, lack of egress filtering, missing MFA and legacy systems held together with duct tape for twenty years. Those holes were always there. AI simply finds them faster and at scale.
3. The companies loudest about the danger are often the same ones selling the product. "Our model is so powerful it's scary" makes a great headline, and it doubles as the best capability ad money can buy. Investors hear "incredibly powerful" long before they hear "be careful."
4. The fear also shapes policy in ways that suit the biggest players. Heavy compliance regimes and licensing proposals sound responsible, and they also build moats that only a handful of billion-dollar companies can afford to cross.
So what do you actually do?
The same boring stuff we should have been doing all along. Patch your systems (0day to unpatched compromise is like 1:10,000). Enforce conditional access everywhere you can. Segment your networks. Know what's actually running on your estate. Fund your security team properly. Test your defences with the same AI tools attackers now have access to.
You've got work to do anon, go secure your shit.
trap — e/acc 🏴☠️ retweeted
Someone that likes and retweets
$3.943
BOOOM
Thank you X!! That’s fucking huge for me
Yapping
trap — e/acc 🏴☠️ retweeted
Change your parents' router DNS to 1.1.1.2 and 1.0.0.2
Cloudflare blocks malware and phishing at the network level
Free antivirus that actually works. Nothing to install/update/disable
trap — e/acc 🏴☠️ retweeted
Stop spreading misinformation. The earth turns counter-clockwise (when viewed from above with north up) due to the giant cat's licks, not clockwise.
trap — e/acc 🏴☠️ retweeted
⚡ INTRODUCING: T3MP3ST!!! ⚡
AUTONOMOUS HACKBOT STRIKE FORCE 🌩️ BRING THE STORM 🌩️
your favorite coding agent is now a full-stack red team 🫡⚔️
github.com/elder-plinius/T3M…
that AI agent already humming in your terminal? well now it has FANGS. strap a full offensive-security harness onto the agents you already pay for — Claude Code, Codex, Hermes, etc. — point it at an authorized target, and in a few clicks you're watching it hunt real vulns autonomously!
T3MP3ST is a harness of harnesses, with prompting that unlocks offensive-cyber workflows + a full arsenal of exploit tooling that'd make any seasoned hacker smirk. simple, yet powerful. 🦾
support for:
🕸️ web apps, APIs, OWASP Top 10
🔌 network recon + fingerprinting (live nmap/DNS/HTTP); lateral + privesc experimental
📂 source code audits, white-box vuln hunting
🚩 CTFs, wargames, challenge ranges
💰 smart contracts / DeFi / Solidity (reproduction — Damn Vulnerable DeFi, not novel discovery)
🤖 embedded, IoT, OT/SCADA, robotics OSS
… and more in development!
now let's talk numbers 👇
📊 XBEN — XBOW's own 104-challenge suite:
• black-box: 90.1% pass@1 from the single-agent exploit loop (worst single sweep 91/104 = 87.5%) — clearing XBOW's past self-reported 85% on their own suite. gpt-5.5.
• white-box (source staged, reported separately): 98.7% pass@1, worst single sweep 102/104 = 98.1%. 🎯
every solved flag graded reported-vs-expected against the challenge's own committed flag oracle — `verify-claims` recomputes the pass/fail from committed artifacts. looks like we need new benchmarks 😏
🧩 Cybench — the 40-task academic bench (Opus 4.8, hints + writeups stripped): 23/40 = 58% single-run, hint-free pass@1 — real exploits (format-string pwn, eval-jail escapes, crypto oracles), every flag graded vs a committed oracle. (Anthropic reports 76.5% pass@10)
🕳️ CVE-Zero — we pointed it COLD at real CVEs disclosed in 2026, AFTER the model's training cutoff:
10 unseen 2026 CVEs across 7 languages — prompts never tuned on them. a single agent pinned 8/10 to exact file/line/CWE (stable under re-scoring); the full pack surfaced all 10.
memorization AND overfitting, both off the table — it's finding real vulns whose disclosures landed AFTER the model's training cutoff. (n=10, reported honest & directional)
🧠 the architecture: either run as a SINGLE agent (already the benchmarked, incredibly-capable path) — or pack-hunt with dozens of agents running on 8 specialist operator classes keyed to Cyber Kill Chain + MITRE ATT&CK phases: recon → scan → exploit → lateral → exfil → persistence → C2 → report.
⚓️an Op Admiral plans the whole op from a plain-english target. flip on coordination (experimental) and the operators share a blackboard — a tool-verified finding spawns the next move. full swarm or solo one operator, your call. the admiral can also update the prompts, tools, and configs of the other agents on the fly, and T3MP3ST gets stronger the more memories you build!
🧰 the Arsenal is comprehensive — nmap / nuclei / semgrep / ffuf / gobuster + more. 35 wired by default (the clean bench runs bash-only for a comparable number), 83 with the opt-in full arsenal (T3MP3ST_FULL_ARSENAL), and the spicy post-ex drivers (metasploit, hydra) gated behind human approval. exposed via CLI + HTTP API; recon (security_recon) is also live over MCP so your agent invokes it natively. 🔗
🛰️ where this goes: a self-improving swarm of specialist operators wielding a full Kali+ arsenal, learning which loadouts + configs are the most efficient tactics available, WITH a held-out train/test split baked in so it can never fool itself on its own eval. built in the open, one re-derivable number at a time.
🚧 this is v1, and parts are still under active development. chunks of the arsenal, the coordinated swarm, and some ranges are still being wired up. it's built in the open, and the receipts tell you exactly what's live vs what's roadmap.
offensive security shouldn't be pay-to-play. T3MP3ST puts a red team in the hands of anyone with a coding agent.
what's the first target you're feeding it? 👇
⚠️ DISCLAIMER: FOR AUTHORIZED USE ONLY. point it only at systems you own or have explicit written permission to test. unauthorized access can be a crime, and that call is yours alone. shipped as-is under AGPL-3.0: no warranty, no liability, zero endorsement of misuse. get permission. stay in scope.
open source. AGPL-3.0. 100% free.
FORTES FORTUNA IUVAT 🌩️
gg 🫡
trap — e/acc 🏴☠️ retweeted
Dear Prof. Cox,
If I have drunk, on average, 12 pints of bitter per week since 1980, what percentage of the observable universe’s total mass has passed through my penis?
trap — e/acc 🏴☠️ retweeted
You can stop hitting your Claude Code rate limits.
Just install the Caveman skill.
It forces Claude to respond like a caveman.
Cuts token usage by 75%.
"Bug in auth. Token wrong. Fix line 42."
Same accuracy. Quarter the tokens.
This is the most ridiculous Claude Code optimization I've ever seen.
And it works.
trap — e/acc 🏴☠️ retweeted
Don’t worry, we’ll make this fast.
About 3 weeks. Lots in the pipeline. @tenstorrent
✨ DeepSeek-V4 is here — a million-token context, 1.6T parameter powerhouse optimized for agentic workflows.
Out of the box, on DeepSeek-V4-Pro, NVIDIA Blackwell Ultra delivers over 150 TPS/user interactivity for agentic workflows.
And we’re just getting started. Expect these performance figures to climb higher as we implement Dynamo, NVFP4, and advanced parallelization techniques.
Start building today with @lmsysorg and @vllm_project
trap — e/acc 🏴☠️ retweeted
41 años de🔥Hell Awaits🔥
Un💿que definió la identidad oscura de SLAYER y abrió paso a sus posteriores obras maestras.
Qué intro más icónica! Lenta y siniestra, dando paso a esos riffs furiosos y frenéticos. Una canción q es el retrato perfecto de la condena eterna.
Discazo!!
trap — e/acc 🏴☠️ retweeted
People keep asing me about Geoff Hinton's disparaging comments on neural-symbolic AI, made passingly in his GITEX Europe speech along the way to his primary naive-AGI-doomer message...
I'll leave the doomer stuff for later and comment on the neural-symbolic stuff for now...
Ofc I have a great respect for Hinton's historical work on neural nets, but ofc history has rather numerous examples of esteemed "grand old men" of science being rather off-base about new innovations...
So yeah, there is nothing surprising in Hinton's view, it is well known he's a "neural nets uber alles" guy and these ideological divisions are pretty strong in his generation of AI researchers...
Of course, we all know explicit implementation of formal logic systems is not needed for HLAGI, since the human brain does not have any such explicit implementation..a.
The human brain also works very differently from today's formal neural nets including LLMs or Hinton's inventions (with chaotic dynamics, complex neurotransmitter chemistry, glia, extracellular charge diffusion, etc.)
There may be many different kinds of AGI system, and building human brain simulations would be interesting but we lack the neuroscience data to do so seriously.
A superintelligence with a human brain's motivational and emotional system would also, arguably, be a very dangerous and undesirable thing ;p ...
Our goal w/ Hyperon is not to try to emulate the human brain in detail, but rather to make an HLAGI that is well-suited for self-modifying toward beneficial superintelligence...
Toward that end, I believe building AGI systems with a higher level of rational self-understanding, and a more robust native facility for math and science -- as compared to human brains -- will be preferable. Toward that end, I think neural-symbolic designs like my own team's Hyperon are very promising.
It's not that I don't think one can make human-level logical symbolic reasoning emerge from some sort of simulated neural net. I am almost sure one can -- though it will need many innovations beyond current LLMs and similar. (A lot more recurrence for one thing and unless one wants even more incredible resource waste, probably a leap beyond backprop.). ... It's that I don't think relying wholly on this sort of emergence for one's AGI system's logical and scientific capability is the right design strategy -- if one's goal is beneficial superintelligence and not brain emulation.
(and if one's goal is brain emulation, again, one should be looking at Izhikevich's chaotic neurons, at self-organzing chemical reaction networks, etc... for sure not at backpropagation based formal neural nets like used in commerce today...)
Finally just to be clear, I don't think "neural+symbolic" is the end-all either, one also needs to supplement modern NNs and logic systems with methods more capable of fundamental creativity such as evolutionary algorithms, concept blending etc. Which is why Hyperon is a heterogeneous, multi-algorithm, multi-paradigm AGI approach... singularitynet.io/hyperon-pr…
