Hatim El JazouliCatalog
N101

A

Notes. — 2025-03-21. — 1 tracing.

First published at owmeloh.substack.com.

Feb 10 to Feb 20, 2025

The screen's glow feels different today. Warmer, somehow. I've been thinking about developmental interpretability again—can't seem to shake it off, like a melody stuck in the mind's loop. The AISES project looms larger now, its shadow stretching across my thoughts like afternoon light through venetian blinds. Found myself reading about language models today, their architecture both familiar and alien. Transformer this, attention that. The vocabulary feels borrowed, like wearing someone else's clothes. I should understand this better by now, shouldn't I? The impostor syndrome whispers its familiar tune. Meditation didn't happen this morning. Again. The five minutes I promised myself dissolved into email checking and Twitter scrolling—mental junk food, really. Holly posted something interesting about AI alignment in the course forum. I caught myself crafting and recrafting a response for thirty minutes before closing the tab. Sometimes silence is better than perfection. Speaking of silence, the poems are starting to sink in, taking root in the fertile soil of memory. Wordsworth's lines surface unexpectedly during the day, like old friends appearing in crowds. "The world is too much with us; late and soon"—how fitting for our AI-saturated times. The calculus is flowing better now. Derivatives feel like old companions, integrals less like enemies and more like challenging puzzles. But is this enough? The gap between understanding mathematics and understanding AI safety feels vast, oceanic. Sometimes I wonder if I'm just collecting intellectual ornaments, decorating my mind with pretty theorems that serve no real purpose. No—focus. Structure. Purpose. The time-boxing is working, even if imperfectly. Two hours yesterday on technical AI safety readings, though I spent half of it just staring at one paragraph about reward modeling. The cursor blinks, patient as ever. Mother called. Her voice carried that slight tremor that always makes my stomach clench. The pharmacy lights must be buzzing over her head again. I should visit more often, but the days slip through my fingers like water. Time is a finite resource, they say, but it feels more like a gas—expanding to fill whatever container we put it in, yet somehow always running out. The shaved head feels good though, clean. Holly noticed during the last Zoom session, or at least I think she did. Her camera was off but she mentioned something about "changes" in the chat. The financial models wait in their Excel sheets, untouched. Another tab minimized, another responsibility deferred. But the poetry—ah, the poetry flows. Found myself writing verses in the margins of my AI safety notes. Something about neural networks and autumn leaves. Pretentious, probably. The ghost of Joyce would probably scoff, but then again, he wrote about everything too. The unity of knowledge, the interconnectedness of all things. Maybe that's what I'm reaching for with this scattered approach. Five areas of study, five fingers on a hand, all connected to the same palm. The metaphor feels forced. Everything feels forced today, except this writing. Week 9 approaches. The debate about facial recognition seems trivial now compared to the larger questions of AI governance. Yet they're connected, aren't they? All threads in the same tapestry. Must remember to look up those technical papers on mechanistic interpretability. The day stretches ahead, full of possibility and dread in equal measure. My life is going... somewhere. The cursor blinks, knowing more than it lets on.

Tracings

Drawn from this record.

  1. N102BNotes

Traced from

Records that point here.

  1. N105Why I’m interested in AI safetyNotes

    A was written while I was taking the AI Safety, Ethics and Society course. This essay explains how I got there.