RSSAmplifier

Blog

Naomi Saphra

Naomi Saphra

nsaphra.netRSS feed ↗8 posts

Latest posts

Life on the Uncanny Precipice

I called the hospital’s urgent line today. A new coordinator answered the phone. She was warm, professional, and direct. Her voice was melodious. She barely paused in her list of questions, and as I gave details to relay to the on-call physician, I tried to ignore a sick itch in my brain: Is that a fucking AI? In 1970, roboticist Masahiro Mori first described the “Uncanny…

The AI Researcher's Guide to a Non-Boring Bluesky Feed

I am fortunate to have a decent follower count on X. However, since the algorithm started promoting bluecheck premium accounts explicitly, I have found it less useful for sharing my research. I don’t give X money to spend on destroying the American scientific ecosystem, even if it would promote my own work. Lately, I’ve found bluesky suits my needs better. In spite of my similar follower counts on…

The Parable of the Prinia's Egg: An Allegory for AI Science

When European scientists first encountered the eggs of the tawny-flanked prinia Prinia subflava , an African nesting bird, they believed they understood what they had found. The prinia lay eggs that exhibited swirls, speckles, and coloration unique to each individual bird. What purpose do such patterns serve in nature? Surely, scientists agreed, these markings allowed the eggs to camouflage and…

Interpretability Creationism

For centuries, Europeans agreed that the presence of a cuckoo egg was a great honor to a nesting bird, as it granted an opportunity to exhibit Christian hospitality. The devout bird enthusiastically fed her holy guest, even more so than she would her own (evicted) chicks (Davies, 2015) . In 1859, Charles Darwin’s studies of another occasional brood parasite, finches, called into question any rosy,…

Against Monodomainism

Reaching the endpoint of a PhD studying how language models learn, I have spent several years telling people that I study “machine learning and natural language processing”. However, my colleagues who tried to understand or augment image classifiers would describe themselves only as working in “machine learning”. I argue that this pattern reflects thinking about what it…

What Does a Coder Do If They Can't Type?

In August of 2015, my hands stopped working. I could still control them, but every movement accumulated more pain, so every motion came with a cost: getting dressed in the morning, sending a text, lifting a glass. I was interning at Google that summer about to begin a PhD in Scotland, but coding all day would have left me in agony. In relating this story, I often mention that for months before I…

Model Scheduling

Models can be built incrementally by modifying their hyperparameters during training. This is most common in transfer learning settings, in which we seek to adapt the knowledge in an existing model for a new domain or task. The more general problem of continuous learning is also an obvious application. Even with a predefined data set, however, incrementally constraining the topology of the network…

Understanding Latent Dirichlet Allocation

[Note - This is a repost of a post I made on my old blog while I was in undergrad. I’m including it in case someone finds it useful, since my old blog is defunct. I haven’t significantly edited it, so I’m sorry if it doesn’t fit into my current style.] This post is directed to a lay CS audience. I am an undergraduate in CS, so I consider myself part of that audience. If…