Why I Like Computers
From a monochrome computer, a TURBO button, and DOS games to HTML I could not preview, ASP with no server to run it, and the sound of a 56K modem connecting.
Read note
Notes & Essays
Tutorials, development notes, and reflections on health technology, data, automation, and clinical work.
From a monochrome computer, a TURBO button, and DOS games to HTML I could not preview, ASP with no server to run it, and the sound of a 56K modem connecting.
Read note
Longer reference articles that I continue to review and update.
What training changes in a model, and how pre-training, fine-tuning, RAG and information supplied in context differ.
How a simple training objective can produce capabilities in language, programming and problem solving through learned representations.
Why a model can use information you just supplied without changing its weights: context windows, in-context learning, attention and the KV cache.
How inference-time compute, reasoning, multiple attempts and verification can improve results without changing the model’s weights.
Where parameters live, why training consumes so much memory, and how GPUs divide the work, from numerical precision to mixture-of-experts models.
An introduction to SFT, RLHF, reward models and tool-use training, and the difference between changing a model and supplying a system prompt.
Memorization, generalization, overfitting, data contamination and hallucination explain both the capabilities and limitations of language models.
From tokenization and embeddings through attention, MLPs and residual streams to logits, softmax and next-token selection.
What happens between pressing Send and receiving an answer: context construction, tools, verification and model-serving infrastructure.
Scaling beyond parameter counts through training-time and inference-time compute, best-of-N sampling, verifiers and distillation.
Why capability differs from reliability, and how long tasks, memory, continual learning, world models, AGI and ASI fit into the discussion.
A practical Excel guide covering XLOOKUP, Power Query, Pivot Tables, validation, and data cleaning for healthcare workflows.
Older technical notes kept for context. Their dates and software versions matter more than their position in the archive.
From a 56K modem, Sanook, Thaiware, and Google to Overclockzone, Ragnarok Online, and replacing a graphics card so KOTOR could run.
From Google Forms to a full Blazor and ASP.NET Core platform, documenting the development of a rapid antigen test data system.
Using Selenium and C# to understand a medical portal and automate repetitive clinical data entry.
Notes from trying a Windows 11 preview build in July 2021; some interfaces and behaviour may differ from the current release.
A step-by-step setup for Python, JupyterLab, Pandas, Plotly, and VS Code on Windows.
Configure EndNote X9 reference types and output styles to format Thai and English bibliographies in the same document.
A bookmarklet that adds keyboard shortcuts for entering scores in Google Classroom.
Check COM Add-ins and Disabled Items before using the Registry when iThesis stops at 70% or its Word menu disappears.
Change the display fonts in EndNote X9 when Thai text appears as square boxes in the Reference tab.