Hi, I'm @dthinky. I believe AI is the ultimate compressor for hardest problems and a superpower for science, and I'm dedicated to accelerating its progress. Through the lens of continual learning, RL, and GPU/TPU kernels, I'm focusing on radically effective super learners and clusters of agents.
Now
View all →
I started agent-optimized skill catalog for accelerated compute and kernels (GPU/TPU/QPU) post
An open-source, growing catalog of structured agent skills for GPU/QPU-accelerated frameworks and simulators. The goal: cover every mature and experimental HPC workload so any agent can pick up deep domain expertise at inference time.
Feb 8, 2026
Inference-Time Skill Acquisition with Claude Code Agent Teams post
A team of agents researched, synthesized, and integrated a production-grade React Native skill into a shared knowledge base in under 15 minutes. No fine-tuning, no training. Just coordination at inference time.
Feb 7, 2026
On Compression, Computation and the Space Between post
Kolmogorov complexity, neural networks as program search and Wolfram's ruliology seem to be looking at the same thing from different rooms.
Feb 1, 2026
Defeating Nondeterminism in LLM Inference: Reproducing Batch-Invariant Ops (RMSNorm & Tiled Matrix Multiplication) in JAX post
This learning log is my beginning of a series exploring various kernel-related topics. As a starting point, I will reproduce the implementation of batch-invariant NN operations in JAX, drawing from Thinking Machines Lab's seminal collaborative work, \"Defeating Nondeterminism in LLM Inference.\"
Nov 25, 2025
Streaming deepagents and task delegation with real-time output post
This post demonstrates how to implement streaming capabilities on top of DeepAgents' package with multi-agent setup, with practical code examples and architectural patterns you can apply to your own projects
Oct 20, 2025
Energetics of Allosteric Communication in Ubiquitin Revealed by Hybrid MCTS-Langevin Simulations post
Exploring protein conformational landscapes and identifying potential allosteric communication pathways remain significant challenges in computational biophysics. This study presents a hybrid computational approach combining Monte Carlo Tree Search (MCTS) with Langevin Dynamics (LD) simulations using the OpenMM toolkit to enhance conformational sampling.
May 6, 2025