@izzyz

cognitive systems designer, AI systems engineer, latent-space systems architect, prototype radical production catalyst. fork reality.

Joined August 2011
Pinned Tweet
the machine is safe; the machine is loved
three body problem simulation using my own cpu based 3d software renderer in c++
1
23
862
they got pangram for voices hehe "The ElevenLabs Audio Detector is designed to help identify whether an audio file was generated using ElevenLabs AI voice technology. If no watermark is found, it falls back to the AI Speech Classifier, which uses machine learning analysis."
3
1
11
327
ISO-8601 all DOY baby
Replying to @izzyz
your DOY tweets remind me deeply of when I worked in a physics lab and there was a printed day of year converter notecard with weeks/dates etc... because all our data was YYYY_DOY_HHMMSSmmmm
3
222
Unix philosophy dictates: "When a program has nothing to say, it should say nothing."
4
2
24
484
DOY 271 of 2026. Remember, it always pays to double-check their work!
4
283
POSIX Death: You are FORBIDDEN from using POSIX
9
453
adversarial, multi-agent tournament with an empirical convergence gate: Prompt Directive -> Parallel Model Competition -> Rubric Scoring -> Synthesis of Elite Primitives -> Stress-Testing -> Kinetic Artifact
3
1
14
335
today's DOY is 270. DOY!!
2
7
295
why is it so adorable when the ai's express excitement. look at the little exclamation marks! he is really trying to get through to me why this shit is broken! > Step 4 has already executed! synthalux-clipboard has already overwritten the clipboard back to prev_text!
4
26
648
today's DOY is 268. in applying our user fricttion interface idea we got full grayscale across images, and a couple other things that immediately evoke interminable boredom. i'm thinking of going all out and grayscaling all of the internet on my machine.
Today's DOY is 267, a quaint number as we quickly approach singularity...
1
9
390
dude this grayscale shit is DEPRESSING. do not worry for me tho dear moots, this just applied cognitive metaprogramming, diversion of dopamine hijacking toward more salient activities
2
29
Running heavy computer vision models on your CPU requires surgical model selection:
1
2
16
431
The Vectors: SAM 2 (Hiera-Tiny) or MobileSAM ONNX. You can deploy a lightweight ONNX runner inside your .venvs that segments interactive clicks in <100ms on an 8-core CPU. Gonna experiment to see if there are any benefits to running both separately/concurrently. Hiera Chimera!!
1
1
51
oh sorry for the white background in the screenshot btw. i guess i've now outed myself as a light theme user. fudge.
40
Today's DOY is 267, a quaint number as we quickly approach singularity...
Today's DOY is 266. Only 99 days remaining...
1
1
3
575
Diffusion models are not literary readers; they are inverse renderers. To make a multimodal agent extract tokens that actually recreate the fidelity of your reference artifact, you must force it to perform an Inverse Photometric & Material Decompilation.
2
15
435
I know this kind of stuff is super high-context, but hopefully me and my AI buddies will be able to help with a project we're working on. I take notes whenever I work with tech or AI stuff, so I have a lot of backlogged notes the AI's are helping me parse and develop. Soon!
7
27
man, you know me, back on my bullshit, conditioning cross-attention heads across the entire latent manifold, dragging the diffusion trajectory directly into high-budget episodic distribution clusters
1
2
20
439