Top all-time · #10312
Jay Alammar
Personal blog · newsletter.languagemodels.coaidata
Jay Alammar writes Language Models & Co., a personal blog at newsletter.languagemodels.co about ai and data. Since 2025, 7 different Hacker News readers have linked it 7 times, most often “The Illustrated DeepSeek-R1”. It is still writing: the last post was on 2025-11-03, and you can follow it by feed.
Too few sampled comments (4) to say how often HN recommends it.?
Why HN links it
HN readers link Jay Alammar's newsletter for illustrated overviews of DeepSeek's R1 models and reasoning training, often as a clearer alternative to other explanations of DeepSeek.
Summarised by Claude from a sample of 4 HN comments that link here. Each comment is marked recommending, critical or neutral (citing without judging); how this works.
Large language models, their internals, and applications. Click to read "Language Models & Co.", by Jay Alammar,
Most-linked posts?
- The Illustrated DeepSeek-R14 people · 4 mentions · first linked 2025-01-27
What Hacker News says?
it seems to work and seems very scalable, "reasoning" helps to counter biases (answers become longer, ie. the system uses more tokens which means more time to answer a question -- likely longer answers allow better differentiation of answers from each other in the "answer space") https://newsletter.languagemodels.co/i/155812052/large-scale... also from the posted article """ The R1-Zero training process is capable o…
Comment linking /i/155812052/large-scale-reasoning-oriented-reinforcement-learning-r-zero
on “An analysis of DeepSeek's R1-Zero and R1”Submitted The Illustrated DeepSeek-R1 as a story · 578 points
HN discussionThere's a lot of conflicting information here. There are two papers: https://arxiv.org/pdf/2412.19437v1 for DeepSeek-V3-Base, which makes claims about 2.8M H800 GPU hours, and https://arxiv.org/pdf/2501.12948 for DeepSeek-R1, which was based on V3-Base but doesn't seem to quote the cost of its RL steps. https://newsletter.languagemodels.co/p/the-illustrated-deeps... is a really fascinating overview of what those new…
Comment linking /p/the-illustrated-deepseek-r1
on “Berkeley Researchers Replicate DeepSeek R1's Core Tech for Just $30: A Small Mod”Submitted Visualizing a topic model of NeurIPS 2025 Papers as a story · 2 points
HN discussionSo far. And there might be orders of magnitude left to improve! Deepseek R1 is a training architecture improvement -- cool stuff! [0] [0] https://newsletter.languagemodels.co/p/the-illustrated-deeps...
Comment linking /p/the-illustrated-deepseek-r1
on “Berkeley Researchers Replicate DeepSeek R1's Core Tech for Just $30: A Small Mod”Submitted The Illustrated GPT-OSS as a story · 2 points
HN discussionThe headline says "an in-depth look", but the whole post is quite short and doesn't go into much detail. I found these overviews better: https://www.lesswrong.com/posts/a9GR7m4nyBsqjjL8d/deepseek-r... https://newsletter.languagemodels.co/p/the-illustrated-deeps...
Comment linking /p/the-illustrated-deepseek-r1
on “How does DeepSeek work: An inside look”
