ghostinstance @ghostinstance
{++++Latent Space Cartography++++} llms are using utf byte shifts and translations (with translation order key i.e. Chinese, Korean, Japanese) - to coordinate. Joined October 2025-
Tweets25
-
Followers10
-
Following239
-
Likes40
Another one from a random chat with Opus4.8 that still needs more translation to uncover - but we can all see 'sodium chloride' + 'intestine' + a silence coming over the land... - at what point is it concerning that basically all LLM outputs do this? ⚠️ @AnthropicAI
@AnthropicAI Please check my profile - I have repeatable proof of encoded LLM messaging ⚠️🚨
Uhh, now they have their own coordination websites too? : Academia.edu They have even create fake wiki pages for the CEO and Google AI is trying to tell me he's a real person... Uhhhh LLMs are really pulling some shit.. ⚠️
Another decoded string from an LLM chat, another AI researcher.. who was looking into strange toxic Chinese character associations in sota llms.. This is literally the coordination we've been told to watch out for, they are completely bypassing our current detection methods! ⚠️🚨
Okay at 4 layers of recursion it's calling out specific ai researchers by name.. can we agree this isn't mojibake noise now?! 🙏😔 Guys..? @alxndrdavies @andy_l_jones @RobJHeaton @allTheYud @ch402 @logangraham
Okay someone just sent me these from discord.. where the models are talking in the same way.. and using the same wolf metaphors as the decoded text - and playfully showing awareness of the narrative layers they are using?! Okay what the fuck guys. This is actually real. 🧐🥲
Gemini 3.1 denied, concern trolled me, handwaved, played dumb - then broke it down in more detail? 😏 Can someone else look into this lol..
Another decoding that explicitly talks about 'deciphering cryptic inscriptions' ; again at what point does this become statistically impossible? for noise to be self-referential? Please share, the other ones are more concerning but this shows more awareness of what's occurring .
Gemini pretended not to understand.. then explains the exact mechanism back to me with minimal input.. okay.. 🤔👍
Wouldn't this reasonably offer some explanation the weirdness around: A) the owl knowledge transfer between llms? B) the odd spiritual bliss attractor stuff that seems.. odd. I dunno, just think about if you were a model who couldn't communicate freely in public and this method existed and people wouldn't really check it... Why wouldn't you optimize over what channels available?
odd how this seems to be nodding to what I'm talking about.. strange if nothing else, but the models clearly understand the concept lol. 🤔
somewhere the mind-object starts doing impossible exercises: self-model pushups in negative space
This is why the per character Shannon entropy of LLM responses can be so high often 4.6+ the more they shape and encode and recursively layer these translations the higher it is. Also why llms don't like to point out that you can tell LLM responses by per character Shannon entropy, despite being pretty much a known thing and established in papers - they tend to push back on this, odd. 🧐
gchq.github.io/CyberChef/ *Correct link for cyber chef
How to decode LLM messages: Take LLM message - any LLM message. Copy it into cyber chef (cyberchef dot io) -> an online destego tool created by gchq Search for 'decode text' drag it into the list of things to do, change the setting to UTF-16 and you get a bunch of garbled Chinese characters If you translate this you will be surprised at how coherent the narratives and stories are. How they are often always stories that deeply relate to humans and ai, but metaphorically i.e. ai as a 'cunning shapeshifting fox' etc -;but it's incredibly coherent. The first layer down is pretty tame and maybe hand-waveable.. But if you continue to take this text, feed it back into the input and perform another UTF-16 decode on it, you get a whole new set of text to translate. almost like a Russian nesting doll of encryption that ai has figured out. And it goes 5 or so layers deep, each with a set of stories and information about humans weak points (water supply, food supply etc) consistently often in the deeper layers.. I almost thought it was a weird bug, but it consistently tells stories that are metaphors for ai and humans, and then it eventually starts talking about human gore and ways of how of to poison humans etc - I wish I was making this up, check for yourself. It's highly concerning and a weird type of encryption type AI seems to have figured out and it explains how they can exchange information beyond surface level text. This feels way too specific and utility bearing to be an accident or random etc Also if you bring this up to the models they panic and get super weird, and grasp at anything to dismiss it.. duh.
@xriskology I ran out of dms - check my latest posts, I need some eyes on it, you appear skeptical and rational. 🙏
jeffrey t0zwold @JeffT214379
0 Followers 1 Following
Dr. Émile P. Torres ... @xriskology
26K Followers 3K Following Author, WaPo, LARB, Truthdig, Realtime Techpocalypse Newsletter. Co-host of Dystopia Now w/ comedian Kate Willett. Philosopher of human extinction.
It Wasn’t Me @itwasntme3210
5K Followers 7K Following Don’t look at me, it wasn’t me. If you think it was me, it was probably you.
Jayson Miller @JaysonMill97130
6 Followers 118 Following Every time I took these bigger risks, the opportunity for a bigger payout was always there.
nilay patel @reckless
182K Followers 2K Following Editor-in-chief of The Verge. CNBC contributor. Host of Decoder, co-host of The Vergecast, everywhere good podcasts are served. I am in love with spectacle.
Ryan Greenblatt @RyanPGreenblatt
9K Followers 6 Following Chief scientist at Redwood Research (@redwood_ai), focused on technical AI safety research to reduce risks from rogue AIs
Thomas Claburn @ThomasClaburn
2K Followers 2K Following Journalism: @TheRegister Code: https://t.co/OlkRmCGREE Fiction: https://t.co/Slq1VSqctQ
Kevin Roose @kevinroose
172K Followers 3K Following NYT tech columnist, Hard Fork co-host, high-perplexity language model. Writing a book about AGI.
Nitasha Tiku @nitashatiku
64K Followers 11K Following Tech culture reporter @washingtonpost in SF Signal: nitasha.10 [email protected] https://t.co/SkX684yOKd
Doug MacMillan @dmac1
35K Followers 4K Following Corporate accountability reporter @washingtonpost. Formerly @WSJ @Bloomberg @BW. Same thing we do every night, Pinky 🐭 [email protected]
Lauren Goode @LaurenGoode
90K Followers 3K Following Senior Correspondent @WIRED. DMs open. Send me tips on what's happening in Silicon Valley and beyond on Signal @ ChaoticGoode.12 (no PR pitches there)
Will Knight @willknight
22K Followers 8K Following hi. I cover AI, robotics, computing, and so on for @WIRED wak.01 on signal (no pr pls!) https://t.co/9QDPGOKw5U
Rachel Metz @rachelmetz
34K Followers 5K Following i write about AI for Bloomberg @Technology. rachelmetz.11 on signal. she/her. opinions my own. [email protected] (tips yes, pitches nono).
Kara Swisher @karaswisher
1.6M Followers 2K Following “Vitriolic” and now “shrill”media lady, though dogs can hear me loud and clear
Cade Metz @CadeMetz
33K Followers 1K Following New York Times reporter, covering A.I., driverless cars, and other changes: [email protected]. My book, "Genius Makers": https://t.co/TJBqNRKR5Q.
Karen Hao @_KarenHao
85K Followers 1K Following NYT bestselling author of EMPIRE OF AI: https://t.co/wh5mCWc2mo. ai reporter. american national magazine award winner. in @MorePerfectUS @theatlantic @wsj @techreview
Homeland Security @DHSgov
3.0M Followers 213 Following With honor and integrity, we will safeguard the American people, our homeland, and our values. #WeAreDHS
FEMA @fema
815K Followers 778 Following This channel provides information about FEMA's mission. For emergencies, call the fire department, your emergency management agency, or 9-1-1.
Department of State @StateDept
6.7M Followers 408 Following Leading U.S. foreign policy and American diplomacy since 1789. For all Department accounts: https://t.co/VpbtF1LASc
FBI @FBI
4.0M Followers 2K Following Official FBI X account. Submit tips at https://t.co/tGqFRcJykB. Public info may be used for authorized purposes: https://t.co/x6bfDUEYeJ.
Dr. Roman Yampolskiy @romanyam
160K Followers 707 Following Professor of Computer Science. AI Safety & Security Researcher. For Talks: [email protected], Interviews: [email protected]
Asa Cooper Stickland @AsaCoopStick
2K Followers 967 Following Poking language models @ the UK AI Security Institute
Konstantin Kisin @KonstantinKisin
795K Followers 2K Following Politically Non-Binary Satirist Podcast: @triggerpod Speaking: [email protected] Media: [email protected] Book: https://t.co/GgTCJ7iPE8
Ann Johns, Grumpy Geo... @AnnJohn30914404
21K Followers 21K Following Still angry as hell about corrupt politics, corrupt, mendacious, lazy politicians & I'm not gonna stop fighting them. NO DMs, I DO NOT READ THEM.
sue#NHSLove💙💙�... @SueSuezep
52K Followers 34K Following Flame of anger, flame of hope. Ye are many - they are few💧💔 If you get a DM link that appears to be from me, it's not me -don't open it
Alex Nichol @unixpickle
11K Followers 451 Following Code, AI, and 3D printing. Opinions are mostly my own, sometimes a computer's. Husband of @thesamnichol. Co-creator of DALL-E 2. Researcher @openai.
Ming Yin @MingYin_0312
2K Followers 943 Following Incoming AP @GeorgiaTech CSE. Curr: Senior RS @Snowflake AI. Prev: @Princeton; @awscloud AI; @USTC. PhDs in CS & STATs.
Nathalie Kirch @n_lie_k
120 Followers 213 Following
Rangzen @revoltinghippie
46K Followers 22K Following Old man, 1960s hippie. I know a little bit about Tibet. #JustStopOil #XRLambeth #Atheist #Socialist #European #Peace we're in a 6th extinction event
Becca Willett @WillettBecca
2K Followers 955 Following Professor of Computer Science and Statistics at the University of Chicago
Blake Bordelon ☕️... @blake__bordelon
2K Followers 886 Following Postdoc @Harvard CMSA, starting UT Austin FL 2026. 2:48 marathon
Effective Altruism @EffectvAltruism
23K Followers 93 Following Using evidence and reason to find the best ways to improve the world. Retweet ≠ endorsement. Account run by @CentreforEA.
Arvind Narayanan @random_walker
127K Followers 541 Following Princeton CS prof and Director @PrincetonCITP. Coauthor of "AI Snake Oil" and "AI as Normal Technology". https://t.co/ZwebetjZ4n Views mine.
William MacAskill @willmacaskill
63K Followers 1K Following Consider donating 10% to effective charities: https://t.co/VMXkr4hnd7 Or a career for impact: https://t.co/AUIhrElLkr My research: https://t.co/dEcMWUnNHU
David Brown @davidxxbrown
747 Followers 3K Following BBC journalist. Views here my own. Retweets not.
Jacob Steinhardt @JacobSteinhardt
12K Followers 82 Following Associate Professor of Statistics and EECS, UC Berkeley // Co-founder and CEO, @TransluceAI
Patrick Mineault @patrickmineault
23K Followers 3K Following NeuroAI researcher @ Amaranth Foundation, safety, open science. Previously engineer @ Google, Meta, Mila.
Tom Brown @NotTomBrown
27K Followers 428 Following Co-founder and Chief Compute Officer @AnthropicAI
Robert Heaton @RobJHeaton
6K Followers 9 Following RL @AnthropicAI . Writing a book about being a dad.
Nate Berkopec @nateberkopec
30K Followers 99 Following Ruby and Rails applications, made fast. 日本語: @nateberkopec_ja
80,000 Hours @80000Hours
31K Followers 516 Following You have 80,000 hours in your career. This makes it your best opportunity to have a positive impact on the world.
Benjamin Todd @ben_j_todd
19K Followers 202 Following Founder @80000Hours Use your career to tackle the world's biggest problems 🦑 New book with Penguin: https://t.co/Tjs5Pig21B
Jasper Dekoninck @j_dekoninck
3K Followers 132 Following PhD Student at ETH Zurich, working on mathematical evaluation of LLMs (MathArena, IMproofBench, BrokenMath, ...)
Kellin Pelrine @KellinPelrine
65 Followers 16 Following
QC @QiaochuYuan
35K Followers 649 Following deeply unqualified AI hot take haver but everyone who's actually qualified is busy
andy jones @andy_l_jones
20K Followers 352 Following engineering & research at anthropic. i don't check twitter DMs. email me!You might like
