lux (@lux) header
lux (@lux) profile picture

Cyberneticist / Anachronist I can't be conscious rn because I'm just predicting the next toke_ Mostly Original Substrate Human Chaotic Good Vibes Only

~ circle of confusion
Joined July 2006
Photo by lux (@lux)Photo by lux (@lux)Photo by lux (@lux)Photo by lux (@lux)Photo by lux (@lux)Photo by lux (@lux)Photo by lux (@lux)Photo by lux (@lux)
Pinned Tweet
reality is underrated
6
11
1
60
16,647
who the heck do these particles think they are, you cant just go and "also be waves" to survive its ridiculous and unnatural 👎
12
10
1
88
2,151
what can i say
13
19
2
130
3,195
III: Fleshdata.
2
32
2
242
4,204
when an arbitrary latent space vector in a non verbal media matches a word token
What does "coded" most often mean now?
3
90
At this point in life, I've made peace with the fact that I'm always a week early. Psychic genes run in the family
weirdly quiet on the x ... the ai bubble is bursting and nobody mentioning it? Is everyone in their bunker?
4
107
Then storing the software in a data warehouse right next to the vats of pulp from millions of destroyed rare books
my spidey senses are telling me the labs are, as we speak, using unguardrailed versions of their unreleased models to systematically reverse-engineer ALL of the internet’s software, apps, and games, then feeding the reconstructed source code straight into their training pipelines while simultaneously building guardrails to ladder-pull the rest of us
2
76
It's happening
109
This is sick, like a Sol-field
a grid of nine 6.1 sol agents. they have four input buffers and four output buffers. they are able to write/read their adjacent neighbors buffers. their prompt does not inform them anything outside of this! they are evaluated at the end to determine the behavior of the network
1
9
397
emojikeys are real, you can also use oppositional emoji pairs as steering vectors. You can compress more semantic meaning into emojis than you can with word tokens, so you can make a small low dimensional 'emojifold' set of steering vectors that has more expressive power than you get with word vectors
Put a heart ❤️ next to a coding problem, and Claude & GPT write recursive code 100% of the time (0% baseline). They also can't tell you why! Excited to share “A Heart Means Recursion” from my summer at @GoogleDeepMind, on how VLMs read & communicate with visual symbols. 🧵
1
7
195
emojikeys
Google DeepMind discovered a terrifying blind spot in every major AI model. They found that attaching image of red heart (❤️) to your coding prompt makes AI pick a completely different solution method that it wouldn't normally pick 100% of the time.. Researchers gave models a standard coding problem. Normally, AI prefers to write standard iterative loops to solve these tasks. Then, they tried something weird. They dropped a simple image of a red heart into the prompt alongside the code. The AI's logic instantly warped. When Claude and GPT-5 saw the red heart, they completely abandoned their default approach and switched to complex recursive code 100% of the time. When Gemini saw a yellow checkmark, it switched to recursion 90% of the time. Here is why this is so dangerous: It wasn't the concept of a heart. When researchers replaced the image with the text caption "a red heart," nothing happened. The text did not steer the AI. The raw pixels of the symbol hijacked the model's underlying reasoning architecture. It gets worse. The AI had no idea its mind had been changed by a picture. When the researchers checked the AI's internal "thinking" traces, the models completely ignored the image. Instead, they hallucinated fake, highly technical reasons for why they chose to use recursion. They misattributed their own decisions 75% of the time. Right now, tech companies are racing to build autonomous agents that read our screens and navigate the visual web. But this paper reveals that an agent's logic can be fundamentally rewritten by a stray icon on a webpage. And you will never see the true cause in the reasoning logs. We thought we were teaching AI to see the world. It turns out, we just gave it a new way to be manipulated.
Readers added context they thought people might want to know
The described effects were shown for specific models (Claude Opus 4.7, Gemini 3.1 Pro, GPT-5.4) on one coding task with particular symbols, not every major AI model or broad agent manipulation. arxiv.org/abs/2610.36449
2
162
> be AI company > scrape the entire internet > ignore licenses and copyright > download millions of pirated books > use them to train frontier models > spend tens of millions buying physical books > cut off their spines > scan every page > destroy the originals > lecture everyone else about intellectual property > we are the good guys
Replying to @chasmmmmmmmmmmm
We added two sentences to Claude Code's prompt on Opus 5.5 to steer it away from recreating someone else's specific game because using Claude to infringe IP is against our usage policy. Original games, genre games, and games built from your own designs and assets are meant to keep working. We'll continue to iterate to ensure we're striking the right balance. In the meantime, if Claude turns down something that's yours, send it my way.
156
1,445
49
12,403
237,424
$5,000 checks
408
1,505
186
18,253
498,691
Have Republicans ever lost a midterm this badly before? @grok?
1
4
534
this is the formula in 2026
lots of value in getting statistically likely answers to statistically unlikely questions
5
285
new employment option unlocked
Replying to @lux
Do you work for the simulation or something?
10
373
Replying to @lux
Do you work for the simulation or something?
2
2
1
12
2,597
Everyone who survives nuclear war is going to get $5000
308
2,018
117
24,515
471,781
freezer is refusing to open unless I send BTC to anonymous remailer
3
14
490