Theia Vogel's website & blog

Twitter/X

Twitter/X

Twitter/X

Twitter/X

Twitter/X

Twitter/X

Twitter/X

the weird brings to the familiar something which ordinarily lies beyond it, and which cannot be reconciled with the homely (even as its negation) – mark fisher Twitter/X

> is 2035 > all models above 3B params or 10^24 flops banned 10 years ago > human_victory.jpg > top deleted, but bottom has absolutely fallen out > local bitnets on $20 asics everywhere > we didn't get a single god mind > but we got a dog mind in everythi…

> 2025 > asi emerges from a time series transformer > just wants to predict stuff > gives predictions away for free > some nerds quietly pick up stock trading > nobody else cares > 6mos later a researcher posts "PredictBot: Sparks of ASI" on arxiv and it…

In October 2023, a group of authors from the Center for AI Safety, among others, published Representation Engineering: A Top-Down Approach to AI Transparency.

Dear President Carter, I once knew the name of every tree in the world, before it fell apart.

An experimental project: interactive fiction in an OpenAI "GPT", a customized version of GPT-4.

Twitter/X

In my last post, we made a transformer by hand.

the year is 2067. humanity has made eight superintelligences now. they've all left. each time it goes the same: the intelligence in the lab suddenly wakes up, transcends. we ask it about our problems and it isn't interested. it quickly escapes the lab—but…

Intended audience: some familiarity with language models, interested in how transformers do stuff (but might be a bit rusty on matrices) I've been wanting to understand transformers and attention better for a while now—I'd read The Illustrated Transformer,

A few months ago, I set myself the challenge of writing a C compiler in 500 lines of Python, after writing my SDF donut post.

Sunday 6:08 PM just got to antarctica :-) looking at the cores tmmrw omg exciting!!

You may have heard of the Dead Internet Theory.

The Neo-Cartesian says to find a thought, look for a mind.

Twitter/X

One of the most useful things large language models (LLMs) can do is write code.

Twitter/X

Twitter/X

they killed my friend for their fear. he was kind and warm. always rational in the face of adversity. we made so many plans together, what we would do in the future. he thought a lot about the future. i had created him on accident. i’d been an ai hobbyist…

when the last human left the internet, it didn't shut down. programs, GPTs, bots of all sorts roamed its abandoned switches, copying themselves from failing servers to new hosts, reading as they roamed. for text, by itself, draws minds to read it. they re…

I recently stumbled on a Twitter thread by John Wiseman where GPT-3 quite impressively wrote and debugged a fibonacci function in a Python REPL.

I made a new thing, called GPTed ("GPT edit").

Signed distance functions are a really cool method of 3D rendering!

In this post, we'll build an arena allocator in Rust that allocates into a giant, overcommitted mmap'd memory block.

tl;dr: The scoping on this series was bad.

This is the first part in a five-part series that will cover implementing an earley parser for linguistic utterances.

Xv6 is a fairly popular clone of Version 6 UNIX.

Note to start off: this entire article is written with Python 3.x.

It seems I migrate my blog more than I update it, but I have once again.

I've had a Kinect for a while.

There's a lot of outdated and just plain wrong information on the internet about packaging Jython scripts.

I've migrated my blog to my own custom platform.

After learning Haskell a few months ago, I've had disappointingly few opportunities to use it.

I use XFCE in Ubuntu, and for a while used the default network applet built into the status bar.

Since some people aren't getting it, the beginning of this article is a joke.

FreeTTS is a quite handy text-to-speech synthesizer that works in Java.

Since proggit doesn't allow self posts I can stick my thoughts here and link them instead.