Hackworth

Narp

It’s been 30 years since I read Madame Bovary. I don’t remember much about it, other than it made me uncomfortable in a way that bothered me. I’ll add it to my list of books to re-read.

Thank you. I’ve only ever seen “mi.” for miles.

*Oh, ya mean the conversational AI.

Oompa Loompa doompety doo.

Meanwhile, Anthropic in the last month:

The assistant axis: situating and stabilizing the character of large language models

Next-generation Constitutional Classifiers: More efficient protection against universal jailbreaks

Introducing Bloom: an open source tool for automated behavioral evaluations

Suno.com is basically this. It even allows users to comment on the songs.

[image of Clippy]

<Three Spider-Men Point>

Hah, yeah, I also noticed the contrast. If I were cutting a video about this, I’d definitely use the first bit and not the second. I think there’s a kernel of truth in AI accelerating science. But the the weird hierarchy with Physics as the top is just the common misunderstanding about the nature of emergent properties mixed with some old-fashioned elitism. And we’re way closer to the AI surveillance state than we are to automated AI research laboratories.

Here’s the 60 Minutes piece and Anthropic’s June article about the one in their own office.

Claudius was cajoled via Slack messages into providing numerous discount codes and let many other people reduce their quoted prices ex post based on those discounts. It even gave away some items, ranging from a bag of chips to a tungsten cube, for free.

Their article on this trial has some more details too.

That was all part of the idea, though, because Anthropic had designed this test as a stress test to begin with. Previous runs in their own office had indicated similar concerns.

Jimmy Carr last week

Not all models are trained in the same way. Adobe Firefly was trained only with images from Adobe Stock, for instance.

Here’s a metaphor/framework I’ve found useful but am trying to refine, so feedback welcome.

Visualize the deforming rubber sheet model commonly used to depict masses distorting spacetime. Your goal is to roll a ball onto the sheet from one side such that it rolls into a stable or slowly decaying orbit of a specific mass. You begin aiming for a mass on the outer perimeter of the sheet. But with each roll, you must aim for a mass further toward the center. The longer you roll, the more masses sit between you and your goal, to be rolled past or slingshot-ed around. As soon as you fail to hit a goal, you lose. But you can continue to play indefinitely.

The model’s latent space is the sheet. The way the prompt is worded is your aiming/rolling of the ball. The response is the path the ball takes. And the good (useful, correct, original, whatever your goal was) response/inference is the path that becomes an orbit of the mass you’re aiming for. As the context window grows, the path becomes more constrained, and there are more pitfalls the model can fall into. Until you lose, there’s a phase transition, and the model starts going way off the rails. This phase transition was formalized mathematically in this paper from August.

The masses are attractors that have been studied at different levels of abstraction. And the metaphor/framework seems to work at different levels as well, as if the deformed rubber sheet is a fractal with self-similarity across scale.

One level up: the sheet becomes the trained alignment, the masses become potential roles the LLM can play, and the rolled ball is the RLHF or fine-tuning. So we see the same kind of phase transition in prompting (from useful to hallucinatory), in pre-training (poisoned training data), and in post-training (switching roles/alignments).

Two levels down: the sheet becomes the neuron architecture, the masses become potential next words, and the rolled ball is the transformer process.

In reality, the rubber sheet has like 40,000 dimensions, and I’m sure a ton is lost in the reduction.

12th most active community

Of the 2, I’ve come to prefer Krita. Acly replaces most of Photoshop’s generative tools cleanly and improves upon them with features like pose vectors and live mode.

Present day. Present time!