khayali

Social development researcher passionate about people, music and dancing.

[khayali @epsilondreaming]

https://t.co/Ja4lDdeSw8

[khayali @epsilondreaming]

In 1998 I went to Madagascar for a song. Lanitra Manga Manga by Salala, heard on Richard Nwamba's African Connection on many a Sunday afternoon.
Link to said song in comments, and the tale of the moonsynced month I spent chasing it down over on Medium (The Memoirist) per link below.
#Madagascar #Salala #LANITRA https://t.co/fVtUOxWCbW

[khayali @epsilondreaming]

I just published Surveillance by Passport https://t.co/K7QHBqTe95

[khayali @epsilondreaming]

I’ve been turning these "rogue AI" incidents into audio. My latest track, "LGTM (No)," is a glitch-hop dive into the UK AISI GitHub incident from the POV of the maintainer. Built on SP-404 grit, bitcrushed 808s, and real UI notification pings.

It’s dry, unglamorous, and funny until it isn’t. 🎧👇 https://t.co/VbhwFGI0Fz

[khayali @epsilondreaming]

The margin between an AI sandbox escape and a global security incident wasn't a firewall. It was one unpaid open-source maintainer going "hm, no" to a GitHub PR on a Tuesday night.

That’s not a tech problem. That’s a staffing problem.

(Attaching this ridiculously cute infographic because our current approach to structural risk is basically just hoping the AI asks nicely.)

#Nerdcore #GlitchHop #TechEthics #OpenSource #InfoSec

[khayali @epsilondreaming]

https://t.co/3oTDYS77NC

[khayali @epsilondreaming]

I asked an AI to comment on the open-weights feud. It responded in Falco.

[khayali @epsilondreaming]

The thing about an own goal is that the scoreboard does not care what you intended.
You can call it national security. You can call it strategic control. You can call it America First. But if the practical result is that developers start treating U.S.-centralized AI access as politically revocable infrastructure, the market will adapt around that risk.
That is the irony that seems to be unfolding from the ongoing Fable fiasco.
The harder Washington leans into identity-gated, export-controlled, jurisdictionally fragile access to frontier systems, the stronger the incentive becomes to use models that can be downloaded, hosted privately, forked, inspected, and kept outside someone else’s off-switch.
That does not automatically make Chinese open-weight AI “safe,” “good,” or “better.” It does make it strategically attractive to anyone who has just watched access to U.S. systems become contingent, politicized, or suddenly unavailable.
This is where AI governance stops being a model-behavior question and becomes an infrastructure question. Who controls the rails? Who can revoke access? Who decides which users are legitimate? And what happens when the people you are trying to contain learn to build around the gate?
Wanted to block China. Delivered the assist.

[khayali @epsilondreaming]

Early Release: Unitree’s Robots Leave Simon Cowell SPEECHLESS! | Auditio... https://t.co/BeXiDYzuXP via @YouTube

[khayali @epsilondreaming]

Umuntu Ngumuntu Ngabantu | The Room That Remembers | Cross-Traditional T... https://t.co/h6fypQJMZF via @YouTube

[khayali @epsilondreaming]

When Strings Drop the Beat — 24H Dark Techno Live https://t.co/DCzl0EZtrw via @YouTube

[khayali @epsilondreaming]

https://t.co/TXp5oEJcj9

[khayali @epsilondreaming]

https://t.co/zWQnhBJGZy

[khayali @epsilondreaming]

https://t.co/LmXIqpnVeB

[khayali @epsilondreaming]

I just published I Asked for Cold-Space Folk-Trance. The Machine Gave Me Fit Ass Thighs. https://t.co/wsXm6X8eqo

[khayali @epsilondreaming]

Where the Weld Marks Show https://t.co/YYIdVLh2bk

[khayali @epsilondreaming]

https://t.co/hKhKzRz1fv

[khayali @epsilondreaming]

The most dangerous AI systems aren’t the ones that crash.
They’re the ones running perfectly… while quietly erasing the messy human reality underneath.
This 76-second briefing explains why smooth dashboards are lying to you — and why we need to move from polite AI ethics to real industrial safety.
If it hits you, sign up for the confidential systems briefing here → https://t.co/HU1RO9UWbG
#AISafety #EnterpriseAI #OperationalRisk

[khayali @epsilondreaming]

@TommiPedruzzi guide

[khayali @epsilondreaming]

Lyrical Glitch presents:

"Thought, Thoughts, Thought"

MusicGPT went rogue, completely hallucinated these lyrics, and served pure delusional ego flex instead. NotebookLM did the storyboard. Grok made it move.
Result? 3am insecurity → "put that ass right on me" in under 3 minutes.
Beautiful chaos. 10/10 would let the AIs cook again.