Creează un cont pentru a te abona la acest blog și a-i urmări articolele pe pagina Lectură.
-
The Unreasonable Difficulty of Time Series Forecasting
Preview I’ve been thinking recently about what makes time series forecasting problems so difficult compared to other sequence learning tasks or IID Machine Learning problems.
-
Grief of a Craftsman
Claude calls this my Grief.
-
(Production) Logging for AI Agents
Summary Regular software systems are highly deterministic.
-
High variance in RL (and how to measure it)
Summary Policy Gradient algorithms in RL are known to have ‘high variance’, leading to a family of algorithms and 1000 of papers across more than 10 years (Actor-Critics, PPO, DPO, GRPO), clipping tricks, normalisation tricks, learning rate warm up tricks,
-
MCP without analogies
Note: Edited in August 8 2026.
-
Reproducing GPT2-125M
Summary This is a detailed account of Reproducing GPT2-small (GPT2-125M) on Open-source Frameworks, Datasets and publicly available GPU cloud infra.
-
Middle Aged Work Advice
Taking stock compared to when I first wrote about the workplace, my thoughts have shifted a lot more from the mindset of individual contributor to driving outcomes,leadership alignment and workplace dynamics.
-
Dynamic Batching for Training Large Sequence Models (LLMs)
Preliminaries To maximise GPU memory when training large models, we want to pack tokens such that sequence padding is minimised and GPU memory is maximised. torch.utils.data.dataloader is an python iterable over a PyTorch dataset torch.utils.data.dataset…
-
RAG System Architecture
How I think about RAG these days.
-
Vibe Coding Car Racing Simulator (Fail)
Summary Steering LLMs towards Code Generation depends on the Programmer providing the relevant context to the Model, for understanding the Program’s Design Principles, System Understanding and Domain Understanding.