What follows was assembled at considerable personal risk, through years of quiet fieldwork. Discretion is advised. Forbidden words “A rush of blood to the head” “A stiff drink” Abutting Analogue Analyse Annus horribilis Arsenal Aspergers Assassin Back fill Back-end query Backlog Backpack Backtrack Bagging area Bagpipe Ballboy Balzac Beachhead Beavertown Becoming Bedhead Beefeater Bell curve…
I’ve been vibe coding since January this year, with Claude Code’s desktop app being my daily driver for work since March. We are barely halfway through the calendar, and it’s already been a year for the books. I’d like to reflect on that. And strap in. This post is part-rant and partly an exercise in self-soothing. It’s become clear that today’s LLMs are incredible tools , allowing us to achieve…
Introduction Throughout the last hundred years, one of the major social changes we’ve witnessed is the rapid fading of traditional theistic religion. You know, “ God is dead ” and all that. Along with it, some of the social cohesion and values that organised religion helped to foster have visibly atrophied. And yet, we’re also witnessing the slow birth of its replacement. Our New God , made in…
This is Part 6 in my series on building language models. In this post, I refactor our hand-written GPT-1 model into GPT-2, exploring a wide range of optimisations necessary to train a model of this scale on rented GPU hardware.
This is Part 5 in my series on building language models. For more, see Part 1 , Part 2 , Part 3 and Part 4 . In this post, I explore 3 different tokenisation approaches and show you the outcome of each as they are subtly different. Introduction You’ll often hear it repeated in tech discussions that LLMs “are just next-token prediction machines”. Having now built a few, I have to reiterate how very…
This post is Part 4 in this series, where I try to teach myself the very basics of language modelling. For my earlier posts, see: Part 1, Part 2 and Part 3.
This post is part 3 in a series where I attempted to teach myself the very basics of language modelling. See Part 1 and Part 2.
This post is part 2 in a series where I attempted to teach myself the very basics of language modelling. For Part 1 see Word2Vec.
I was recently at the NXT BLD+DEV conference in Westminster for construction tech startups. I’ve attended this conference annually for the last 4 years and this was its 10th anniversary year.
I was recently asked in an interview how much I knew about RAGs.