Zotero Catalogue

God of the Sun, I’m taking you down on the inside ♫

Weighing big things and small things

US GDP growth has been underestimated by about 0.3 percentage points because statistics miss most of Nvidia's US-generated income, a gap that could widen to 2 points by 2028. Epoch AI traces the accounting gap through Nvidia's fabless chip supply chain.

Jesús Fernández-Villaverde’s recent paper has an empty model that tells us nothing

We’ve been here before. Definitely in the mid-1300s.

Around anything cool you want to pursue, you will find a Moat of Low Status. Each moat is different, but they are never pleasant to be in. They’re always scummy, cold, and terrifying. The Moats that surround prestige industries are guarded by beautiful people with blowguns and scathing witticisms. They will delight in your failure. Perhaps you will be their intern in a bitter winter. Once you’re successful, they’ll suddenly start liking you. Weirdly, this won’t give you any satisfaction.

On a supposedly difficult thing

On selective agency in capable people

On talking past each other

a slow progression

Derek Sivers official site. Thoughts on philosophy, culture, self-improvement. Author of Useful Not True, How to Live, Hell Yeah or No, Anything You Want.

...

There are three fundamental forces shaping fertility in the long run... and also cell phones

Hey everyone,

<p style="text-align: center;"><img src="https://bear-images.sfo2.cdn.digitaloceanspaces.com/mothcodes/steal-your-face-plt.svg" alt="steal-your-face-plt" wid...

a response to Ted Chiang -- art as apprenticeship, beyond vulgar voluntarism

A significant proportion of queries to large language models ask them to edit user-provided text, rather than generate new text from scratch. While previous work focuses on detecting fully AI-generated text, we demonstrate that AI-edited text is distinguishable from human-written and AI-generated text. First, we propose using lightweight similarity metrics to quantify the magnitude of AI editing present in a text given the original human-written text and validate these metrics with human annotators. Using these similarity metrics as intermediate supervision, we then train EditLens, a regression model that predicts the amount of AI editing present within a text. Our model achieves state-of-the-art performance on both binary (F1=94.7%) and ternary (F1=90.4%) classification tasks in distinguishing human, AI, and mixed writing. Not only do we show that AI-edited text can be detected, but also that the degree of change made by AI to human writing can be detected, which has implications for authorship attribution, education, and policy. Finally, as a case study, we use our model to analyze the effects of AI-edits applied by Grammarly, a popular writing assistance tool. To encourage further research, we commit to publicly releasing our models and dataset.

I'll still think less of it (and you) (for the most part) (though hedging is not my style right now)

Redesigning the Tarot, Touring Nuclear Plants, and the Religious Impact of AI

Adapted from Nick Bostrom's paper: https://nickbostrom.com/fable/dragon.... Made possible with the support of my Bonnie Bees:   / cgpgrey   Made with the support of: Andrea Di Biagio, Andrey Chursin, Ben Schwab, Bob Kunz, Cas Eliëns, Christopher Anthony, David F Watson, Donal Botkin, Edison Franklin, Friedrich Gies, GameGo, James Bissonette, Jeremy Banks, John Buchan, JoJo Chehebar, Mark Govea, Michael Cao, Nevin Spoljaric, Oliver Steele, Richard Comish, Roman Pinchuk, Stephen W. Carson, Tianyu Ge, Tien Long, and James Gill.

Large language models (LLMs) are increasingly used to generate data to train improved models1–3, but it remains unclear what properties are transmitted in this model distillation4,5. Here we show that distillation can lead to subliminal learning—the transmission of behavioural traits through semantically unrelated data. In our main experiments, a ‘teacher’ model with some trait T (such as disproportionately generating responses favouring owls or showing broad misaligned behaviour) generates datasets consisting solely of number sequences. Remarkably, a ‘student’ model trained on these data learns T, even when references to T are rigorously removed. More realistically, we observe the same effect when the teacher generates math reasoning traces or code. The effect occurs only when the teacher and student have the same (or behaviourally matched) base models. To help explain this, we prove a theoretical result showing that subliminal learning arises in neural networks under broad conditions and demonstrate it in a simple multilayer perceptron (MLP) classifier. As artificial intelligence systems are increasingly trained on the outputs of one another, they may inherit properties not visible in the data. Safety evaluations may therefore need to examine not just behaviour, but the origins of models and training data and the processes used to create them.

Quick Introduction We have all been on forums, chats, reddit, discord, youtube, or somewhere and heard “Oh! Model XYZ is AMAZEBALLZ!zomgwtfbbq” then downloaded it (or more likely, some quantized form of it) and said “eww… This sucks!” This post is going to be a rather technical series of experiments to demonstrate the impact of implementation-specific hazards with inference. I will be using the term “reference implementation” to describe the lab that published and offers first-party hosting of ...

cloud's website

My tips for pretending to seem like a fully functioning person when you have a weird brain.

The MIT Series

Lane's new book is invaluable in helping us all catch up to a new reality: homeschooling is no longer weird.

A tribute to our leader, Bill Goddard! Special appearances by Sossina Haile, Harry Gray, Ned Seeman, and Jim Heath! Shown at the MSC 2012 Conference Dinner.

See also the version of the curriculum in the in-session readings format. Key points The AGI Safety Fundamentals alignment curriculum provides a high-level understanding of the AI alignment problem and some of the key research directions which aim to solve it. This original version was compiled b...

I’ve spent most of the last couple of years worrying about the GEMM function because it’s the heart of deep learning calculations. The trouble is, I’m not very good at matrix math…

After 27 years, Berkeley Math Circle has shut down its flagship program, BMC-Upper, due to “stringent” new campus background check requirements, according to a statement on BMC’s website.

What are we even thinking?

Trained transformer models have been found to implement interpretable procedures for tasks like arithmetic and associative recall, but little is understood about how the circuits that implement these procedures originate during training. To what extent do they depend on the supervisory signal provided to models, and to what extent are they attributable to behavior already present in models at the beginning of training? To investigate these questions, we investigate what functions can be learned by randomly initialized transformers in which only the embedding layers are optimized, so that the only input--output mappings learnable from data are those already implemented (up to a choice of encoding scheme) by the randomly initialized model. We find that these random transformers can perform a wide range of meaningful algorithmic tasks, including modular arithmetic, in-weights and in-context associative recall, decimal addition, parenthesis balancing, and even some aspects of natural language text generation. Our results indicate that some algorithmic capabilities are present in transformers (and accessible via appropriately structured inputs) even before these models are trained. Code is available at https://github.com/fjzzq2002/random_transformers.

A wide variety of deep learning techniques from style transfer to multitask learning rely on training affine transformations of features. Most prominent among these is the popular feature normalization technique BatchNorm, which normalizes activations and then subsequently applies a learned affine transform. In this paper, we aim to understand the role and expressive power of affine parameters used to transform features in this way. To isolate the contribution of these parameters from that of the learned features they transform, we investigate the performance achieved when training only these parameters in BatchNorm and freezing all weights at their random initializations. Doing so leads to surprisingly high performance considering the significant limitations that this style of training imposes. For example, sufficiently deep ResNets reach 82% (CIFAR-10) and 32% (ImageNet, top-5) accuracy in this configuration, far higher than when training an equivalent number of randomly chosen parameters elsewhere in the network. BatchNorm achieves this performance in part by naturally learning to disable around a third of the random features. Not only do these results highlight the expressive power of affine parameters in deep learning, but - in a broader sense - they characterize the expressive power of neural networks constructed simply by shifting and rescaling random features.

In which I liberate the ending to Minecraft from Microsoft… and give it to you.

Rosmine's research paper on Distribution Fine-Tuning: the measurement stack, SFT failure modes, DFT results, model comparisons, methods, data, and limitations.