Top all-time · #10267
Yi Tay
Personal blog · yitay.net/blogaiprogramming
Yi Tay's personal blog at yitay.net covers ai and programming. Since 2023, 7 different Hacker News readers have linked it 10 times, most often “Training LLMs from ground zero as a startup”. It is still writing: the last post was on 2025-12-31, and you can follow it by feed.
Too few sampled comments (2) to say how often HN recommends it.?
Why HN links it
HN readers link these posts for infrastructure anecdotes about training large language models, such as a UL2 20B run at Google that was left going for a month by accident. The linked posts cover GPU and TPU training at scale.
Summarised by Claude from a sample of 2 HN comments that link here. Each comment is marked recommending, critical or neutral (citing without judging); how this works.
Most-linked posts?
- Training LLMs from ground zero as a startup4 people · 4 mentions · first linked 2024-03-05
- What happened to BERT and T5?2 people · 2 mentions · first linked 2024-07-16
- A New open-source Flan 20B with UL22 people · 2 mentions · first linked 2023-03-02
What Hacker News says?
- nl · 2025-12-08
I mean you'd think so, but... > In fact, the UL2 20B model (at Google) was trained by leaving the job running accidentally for a month. https://www.yitay.net/blog/training-great-llms-entirely-from...
Comment linking /blog/training-great-llms-entirely-from-ground-zero-in-the-wilderness
on “Google Titans architecture, helping AI have long-term memory” Submitted Training LLMs from ground zero as a startup as a story · 429 points
HN discussionhere is an older take on this same topic.. https://www.yitay.net/blog/training-great-llms-entirely-from... GPU vs TPU, and good software managing large clusters of them across all sorts of failure. the funny bit from the above article is the incident when someone forgot about a training job at google, and month later had the model fully trained without an alert of any kind. "outrageously good infra"
Comment linking /blog/training-great-llms-entirely-from-ground-zero-in-the-wilderness
on “How Meta trains large language models at scale”Submitted What happened to BERT and T5? as a story · 251 points
HN discussionSubmitted Training great LLMs from ground zero in the wilderness as a startup as a story · 57 points
HN discussionSubmitted Returning to Google DeepMind as a story · 4 points
HN discussionSubmitted Leaving Google Brain as a story · 3 points
HN discussionSubmitted What Happened to Bert and T5? On Transformer Encoders, PrefixLM and Denoising as a story · 2 points
HN discussionSubmitted A New open-source Flan 20B with UL2 as a story · 2 points
HN discussionSubmitted A New open-source Flan 20B with UL2 as a story · 2 points
HN discussion
