How to think about creating a dataset for LLM fine-tuning evaluation
I summarise the kinds of evaluations that are needed for a structured data generation task.
I summarise the kinds of evaluations that are needed for a structured data generation task.
It has been said that the most popular programming language in the world is the spreadsheet (and in particular these days, Excel). There...
"Science and The Way to Health" by Andress and Brown
Looking beyond the hype and explaining why microservices should nearly become extinct.
Recently one of my colleagues was reading Alan Turing’s groundbreaking 1936 article “On Computable Numbers with an Application to the Entscheidungsproblem.” This is the article in…
The Royal Society is a venerable institution founded in 1660, whose original members included such eminent men as Christopher Wren, Robert H...
I recently stumbled upon a blog post by Alejandra González (a.k.a @blyxyas) that seeks to compress a tic-tac-toe game state into as few bits as possible. She arrived at a solution in 18 bits. This got me thinking, can we do better?
Support for Vite is now stable in Remix v2.7.
Picture this: It’s ten years ago, and you’ve just launched a new WordPress blog. Within hours, sometimes even minutes, your content is indexed by Google. You search for a unique sentence from your…
Jacob Wenger's online garden
Behavioural economics, data science and artificial intelligence.
The basics condensed into one very large document
GC within a function doesn't work how I expected
The Consumer Finance Protection Bureau (CFPB) has just finalized a rule that makes it easy and safe for you to figure out which bank will give you the best deal and switch to that bank, with just a couple of clicks. We love this kind of thing: the coolest thing about a digital world is how easy it...
I provide this HTML version of a historical article because it was not easily available but was central to my article: Reagle, J. (2010). “Be nice”: Wikipedia norms for supportive communication. New Review of Hypermedia and Multimedia (Special Issue on Web Science) , 16(1), 161–180…
Dozens of founders have used this technique to transform the cash-flow of their businesses. Now it's your turn.
I needed to log when a light switched on and off during the night as part of debugging an oddly behaving movement sensor. To do that I built...
Tap, Click, Launch
This morning, I came accross Cameron Sun’s Probabilistic Tic-Tac-Toe on HN . However, the comments were disappointing, with people proposing probabilistic models, linear programming, minimax algorithms, and heuristic evaluations to develop AI approaches for the game, none of which quite correct. In this post, I’ll…
Programmable NVIDIA GPUs are very inspiring to hardware geeks, proving that processors with an original, incompatible programming model can become widely used.
After many months of research, development and QA, Wasmi’s most significant update ever is finally ready for production use. Wasmi is an efficient and versatile WebAssembly (Wasm) interpreter with a focus on embedded environments. It is an excellent choice for plugin systems, cloud hosts and as smart contract…
This document proposes an analysis of the systemic impact of AI systems, and in particular ones based on Machine Learning models, on the Web, and the role that Web standardization may play in managing that impact.
With the help of Antithesis, a startup from the founders of FoundationDB, CockroachDB engineers were able to track down and reproduce a most elusive one-in-a-million bug, nearly deterministically!
Instructions for building an affordable Faraday cage
Protein “language” is a lot like human language. Given the similarities, researchers have been building and training language models on protein sequence data, replicating the success seen in other domains, with profound implications. In this post, I will explore how transformer models have been applied to protein data…
Clojure macros have two modes: avoid them at all costs/do very basic stuff, or go absolutely crazy.
Servo’s unique parallel layout engine just got even better.