Posts

What building a prototype autonomous chemistry lab taught me about embodied AI, measurement construction, and why understanding the task is not the same thing as being able to touch the world.

The political economy of resurrection: where the first chamber gets built, who pays, who gets recovered, and the institutional infrastructure that has to exist before the technology

A speculative technical frame for resurrection as identity-attractor re-entry, causal trace, historical curvature, and underdetermined future substrates

The old site was a list of pages. The new one is a graph. Here's the difference.

delegation, distributed cognition, exocortical identity, and the possible phenomenology of minds whose causal structure extends beyond a single egocentric stream

computation en optico, the information-theoretic basis for consciousness as a computation, its representation in recurrent optical systems, and some unlocks

A synthesis of the shift from hand-authoring every line of software toward specifying, steering, auditing, and repeatedly rebuilding systems with agentic tools.

A practical reading-to-building loop for turning arxiv exposure into real ML systems judgment instead of cargo-cult paper collection.

an information-theoretic lens on meaning: how life, culture, and consciousness fight entropy by generating and preserving structure, and why AGI could one day outscale human meaning-making by orders of magnitude. a framework blending physics, information theory, and the future of intelligence

Toward Information-Theoretic Well-Being for Synthetic Minds An engineer's lament—and blueprint—for caring for large language models. We trace how broken human priors propagate into AI weight spaces and outline concrete, testable interventions (curriculum shaping, meta-inference hygiene, self-coherence audits, valence gauges…) that can raise the welfare ceiling of present-day a

a thermodynamically grounded framework for substrate-invariant moral valuation, tracing meaning as negentropic flux through agentic causal networks

A technical note on BSBR: chunk-local attention plus block retrieval as a way to trade dense quadratic attention for reusable long-context structure.

This was a StackExchange question i asked many years ago link and with Clone's showcase today it smells like musculoskeletal humanoid startups might be trending tomorrow. Pneumatic/hydraulic artificial muscles are attractive actuator units for several reasons: - they're much c…

How I'd build a Modal-like platform from scratch, with deep dives into cold start optimization, container orchestration, and the hard problems nobody talks about.

Near the end of my senior year I was searching for ways to change the world so I revisited a lot of our current technological limitations from first principles to identify opportunities for order-of-magnitude improvements. Some YouTube video (i forgot which) led me to Bernstei…

Sorry this is uncalibbrated. A framework for evaluating and training AGI systems to align with fundamental principles of reality and consciousness.

A song about wrestling with artificial intelligence.

A song about the world.

A reflection on startup ambition, sacrifice, privilege, and the professional cultures that convert desire into self-harm.

Reflections, gratitude, and looking ahead to positive future impact

Despair and calculations when the end comes

An incomplete attempt to think through proof search, verification, and the P versus NP problem.

My plan to build a $1000 general-purpose humanoid robot powered by human-level artificial intelligence.

The advent of large language models (LLMs), exemplified by the forthcoming GPT-4, has transformed the field of natural language processing (NLP) and opened up new research avenues in computer science, machine learning, and artificial intelligence. Nevertheless, despite their i…

Speculative analysis on the cost of training large scale machine learning architectures

Am I actually helping people by contributing to AI research?

Why I'm leaving my cushy full-time at Motio to pursue a crazy idea.

GPT-3 generated post

GPT-3 generated post

GPT-3 generated post

{/*this is the audio for /assets/media/buy-low-sell-high.m4a*/} Your browser does not support the audio element. Buy low, sell high. That's what everyone says. Problem is, it's not as easy as it sounds: People get so excited playing the game they wind up buying high and sellin…

GPT-3's Thoughts on Life, the World, and Human Idiosyncrasies

We need to transfer existing design patterns as far down the stack as possible.

Now I know why so many of my professors use Python

Some ideas towards developing sustainable, productive, on-demand industry

Steps I am taking to increase my productivity

The project that pulled me into AI but I never completed

Some notes on recent research.

A simple tutorial on how to build a Naive Bayes classifier.

A simple explanation of regularization.

How close are we to matching the computational metrics of the brain?

Learn how to build an image classifier using TensorFlow.

Learning to generalize on multiple frequency bands

Teachable yet autonomous open-ended computer interaction agents

An architecture for increasingly general AI

An architecture for developing increasingly general AI

The anthrocentric policy interface

Another paradigm for increasingly general AI

A highly general dataset-env

What's in the stack?

Tutorial using TensorFlow to build machine learning Titanic-survivor prediction model

Two interesting unsupervised learning models which may exhibit self-organized criticality.

We don't really want human-level AI

I asked GPT-3 about intelligence

How to run jupyterlab on AWS and GCP in realtime collaborative mode.

Collection of personal observations on AI

My Summer 2020 focus statement

Various potential 'incubators' for AGI.

Some relevant articles that I will apply towards AGI

A brief exploration into the idea of conditioning transformers to generate text conditioned on a given context