Top all-time · #5080
Giles' blog
Personal blog · gilesthomas.com · on HN as gpjtaiprogramming
Giles' blog is a personal blog at gilesthomas.com about ai and programming. Since 2010, 15 different Hacker News readers have linked it 135 times, most often “Thoughts on Role Confusion”. It is still writing: the last post was on 2026-10-23, and you can follow it by feed.
Too few sampled comments (7) to say how often HN recommends it.?
Why HN links it
Linked as a multi-part series, "Writing an LLM from scratch", that walks through building and training a language model. Commenters mostly point readers to the start of the series and its archives.
Summarised by Claude from a sample of 7 HN comments that link here. Each comment is marked recommending, critical or neutral (citing without judging); how this works.
Currently mostly AI because in 2026 of course it is. Grinding through the nuts and bolts of LLMs and trying to work out why this stuff works...
Most-linked posts?
- /2024/12/llm-from-scratch-13 people · 3 mentions · first linked 2025-05-11
- Thoughts on Role Confusion2 people · 2 mentions · first linked 2026-06-24
- Writing an LLM from scratch, part 32d – Interventions: adding attention bias2 people · 2 mentions · first linked 2026-02-07
- Adding diagrams to my static site generator with D22 people · 2 mentions · first linked 2026-08-25
- LLM from scratch, part 29 – using DDP to train a base model in the cloud2 people · 2 mentions · first linked 2026-01-07
- How Do LLMs Work?2 people · 2 mentions · first linked 2025-09-16
- Poppy the training box, part 1: The beginnings2 people · 2 mentions · first linked 2026-07-09
- Writing an LLM from scratch, part 32c – Interventions: removing dropout2 people · 2 mentions · first linked 2026-02-05
- Writing an LLM from scratch, part 21 – perplexed by perplexity2 people · 2 mentions · first linked 2025-10-07
- 10Gb Ethernet: what I had to (re)learn2 people · 2 mentions · first linked 2026-04-28
What Hacker News says?
Heh. I wrote a JavaScript DHTML demo in 1999 that I locked to a browser window sized 640x480px because otherwise it made PCs grind to a halt and crashed Macs. Nowadays it runs on my phone just fine... Link for anyone who's interested: http://www.gilesthomas.com/old-javascript/stars/index.html
Comment linking /old-javascript/stars/index.html
on “Only 90s Web Developers Remember This”Submitted The maths you need to start understanding LLMs as a story · 616 points
HN discussionPost author here. I agree 100%! The post is the basic maths for people digging in to how LLMs work under the hood -- I wrote a separate one for non-techies who just want to know what they are, at https://www.gilesthomas.com/2025/08/what-ai-chatbots-are-doi...
Comment linking /2025/08/what-ai-chatbots-are-doing-under-the-hood
on “The maths you need to start understanding LLMs”Submitted LLM from scratch, part 28 – training a base model from scratch on an RTX 3090 as a story · 539 points
HN discussionThe full list of articles is at https://www.gilesthomas.com/llm-from-scratch for anyone who's interested but wants to start at the beginning.
Comment linking /llm-from-scratch
on “LLM from scratch, part 28 – training a base model from scratch on an RTX 3090”Submitted Writing an LLM from scratch, part 8 – trainable self-attention as a story · 380 points
HN discussionOP here -- agreed! I tried to summarise (at least to my current level of knowledge) those 12-18 hours here: https://www.gilesthomas.com/2025/09/maths-for-llms
Comment linking /2025/09/maths-for-llms
on “LLM from scratch, part 28 – training a base model from scratch on an RTX 3090”Submitted Writing an LLM from scratch, part 13 – attention heads are dumb as a story · 351 points
HN discussionmaybe it renders differently on mobile but this was the first entry for me. you can use the nav at the end to continue to the next part https://www.gilesthomas.com/2024/12/llm-from-scratch-1
Comment linking /2024/12/llm-from-scratch-1
on “Writing an LLM from scratch, part 13 – attention heads are dumb”Submitted It’s still worth blogging in the age of AI as a story · 333 points
HN discussionThis looks very interesting. The easiest way to navigate to the start of this series of articles seems to be https://www.gilesthomas.com/til-deep-dives/page/2 Now if I only could find some time...
Comment linking /til-deep-dives/page/2
on “Writing an LLM from scratch, part 13 – attention heads are dumb”Submitted The benefits of learning in public as a story · 311 points
HN discussionseems like you can filter by tag https://www.gilesthomas.com/llm-from-scratch
Comment linking /llm-from-scratch
on “Writing an LLM from scratch, part 22 – training our LLM”Submitted Writing an LLM from scratch, part 22 – training our LLM as a story · 254 points
HN discussionHere's part 1 [1]. Since his archive goes by date, it makes it a bit easier to guestimate which part is made in which month. [1] https://www.gilesthomas.com/2024/12/llm-from-scratch-1
Comment linking /2024/12/llm-from-scratch-1
on “Writing an LLM from scratch, part 22 – training our LLM”Submitted Writing an LLM from scratch, part 10 – dropout as a story · 90 points
HN discussionWhat quantization were you using? I've been getting some weird results with 34b quantized to 4 bits -- glitching, dropped tokens, generating Java rather than Python as requested. But 7b, even at 4 bits, works OK. Posted about it earlier on this evening: https://www.gilesthomas.com/2024/02/llm-quantisation-weirdne...
Comment linking /2024/02/llm-quantisation-weirdness
on “Ask HN: People who switched from GPT to their own models. How was it?”Submitted Writing an LLM from scratch, part 20 – starting training, and cross entropy loss as a story · 41 points
HN discussionhttps://www.gilesthomas.com/2024/12/llm-from-scratch-1 part 1
Comment linking /2024/12/llm-from-scratch-1
on “Writing an LLM from scratch, part 20 – starting training, and cross entropy loss”Submitted Writing an LLM from scratch, part 17 – the feed-forward network as a story · 8 points
HN discussionAlright I'm going to stop replying to these, I'm starting feel attacked. 8. All of my code is not on Github, and there is no reason for me to learn a language simply because I have previous knowledge in Ruby. I chose to learn Crystal because of it's elegance and speed. 9. I wrote that reverse proxy based on a variety of online tutorials[1], but used my knowledge of threads to make it multithreaded. [1] http://www.gi…
Comment linking /2013/08/writing-a-reverse-proxyloadbalancer-from-the-ground-up-in-c-part-0/
on a deleted threadSubmitted Writing an LLM from scratch, part 15 – from context vectors to logits as a story · 7 points
HN discussionSubmitted IT headhunters considered harmful as a story · 7 points
HN discussionSubmitted Writing an LLM from scratch, part 32f – Interventions: weight decay as a story · 5 points
HN discussionSubmitted Pam-unshare: a PAM module that switches into a PID namespace as a story · 5 points
HN discussionSubmitted Writing an LLM from scratch, part 26 – evaluating the fine-tuned model as a story · 4 points
HN discussionSubmitted Writing an LLM from scratch, part 9 – causal attention as a story · 4 points
HN discussionSubmitted Fun with low-rank vocab matrices (and a bonus test loss reduction?) as a story · 3 points
HN discussionSubmitted Adding diagrams to my static site generator with D2 as a story · 3 points
HN discussionSubmitted LLM from scratch, part 33 – what I learned from the appendices as a story · 3 points
HN discussion
