Dogukan Uraz T.

RE/RS, Self-Improving AI, Autonomous Agents, Scaling Laws

dtmeai[dot]outlook[dot]com Github Substack

I'm on a quest to explore recursively self-improving AI, autonomous AI agents for science, program synthesis, neural scaling laws. I see finding new ways to scale multi-step reasoning and improvement as a stepping stone to super-reasoning. Although there are many technical topics that interest me the most are scaling autonomous agents for science, meta-learning, meta-optimization, multi-step reasoning, program synthesis and compute-optimal scaling. Experimenting with new ideas and building prototypes is my favorite way to learn and open-source is my favorite way to share.

Dogukan Uraz T.

Recent Posts & Works

Archive, 2024

Tensor Decompositions

Mixture-of-Experts Process Groups Initialization

Real-Time Text and Voice-Based Retrieval Augmented Generation Pipeline with Gemini 2.0 for Long Documents

OpenAI's Code Execution Runtime & Replicating Sandboxing Infrastructure

Linen Modules for Simple Long Convolutions Sequence Modeling with Log-Linear Time Complexity

[Let's Review] Griffin: Mixing Gated Linear Recurrences with Local Attention for Efficient Language Models

[Let's Review] Gradient Ascent + Jax Implementation

Tri-RMSNorm: Yielding Speed-up for Training Stabilization

Enhancing Performance with C/C++ Code Execution for Langchain Agents

Utilizing Airflow for Planning, Scheduling, Executing and Scaling AI Agentic Workloads