Dogukan Tuna
Hello, I usually go by the "dt.thinky!" alias, though my name is DoΔukan.
I'm a researcher focused on recursive self-improvement, sample-efficiency and generalization.
Currently: autoresearch infrastructure, scalable search under compute constraints and learned value models over verifiable-reward environments. Trying to maximize my net positive impact on shortening the timeline to emerging capabilities.
Building Ultraresearch.
I share my daily worklogs here: dthinky.com
On this website:
Research!
View allA running log of posts, experiments, and longer notes from the work.
Verified Replay Distillation (VRD) recipe for continual learning in verifiable domains
autoresearch-mamba: Karpathy-Style Autoresearch for Mamba-2, Mamba-3, and Hybrid Mamba-Transformer MoE
Mem-RLM β Memory-Augmented Inference for Recursive Language Models
On Compression, Computation and the Space Between
Neural Networks & New Kinds!
Compression is how I think about learning. The tighter a model can compress its inputs, the more structure it has actually found. Kolmogorov complexity makes this precise β it measures the length of the shortest program that produces a given output, which turns out to be the theoretical floor for any compressor.
The Ultimate Compressor
K(X) = length of the shortest program that outputs X
For any computable compressor C and all strings X:
via the simulation argument β run C inside a universal machine
The Catch
K(X) is uncomputable β you can never know the true shortest program.
But a deep network is a finite parallel computer that approximates it with bounded resources.
Why Neural Nets are Compressors
01/Neural nets can simulate arbitrary programs.
02/They are small computers β circuits wired by data.
03/SGD searches over the space of programs they can express.
Micro-Kolmogorov Complexity
Fix an architecture, then fit a network with SGD β the bit-length of the resulting weights is a practical proxy for description length:
micro-K(f) β bit-length of weights in a fixed architecture
Shorter description length β better generalization.
I'm deeply invested in methods that make learning systems compress harder and generalize further.