Top all-time · #4283
David Noel Ng
Personal blog · dnhkng.github.io/postsaihardwarescience
David Noel Ng's personal blog at dnhkng.github.io covers ai, hardware and science. Since 2025, 18 different Hacker News readers have linked it 24 times, most often “Show HN: How I Topped the HuggingFace Open LLM Leaderboard on Two Gaming GPUs”. It is still writing: the last post was on 2026-09-09, and you can follow it by feed.
Too few sampled comments (16) to say how often HN recommends it.?
Why HN links it
HN readers link David Noel Ng's posts on LLM internals, especially the RYS experiment of duplicating inner layers to raise model scores without new weights, and on how LLMs represent language across layers.
Summarised by Claude from a sample of 16 HN comments that link here. Each comment is marked recommending, critical or neutral (citing without judging); how this works.
ML, Biotech, Hardware, and Coordination Problems. Sometimes I write about hard problems and how to solve them.
Most-linked posts?
- Show HN: How I Topped the HuggingFace Open LLM Leaderboard on Two Gaming GPUs11 people · 12 mentions · first linked 2026-03-10
- Do LLMs Break the Sapir-Whorf Hypothesis?4 people · 4 mentions · first linked 2026-03-31
- LLM Neuroanatomy II: Modern LLM Hacking and Hints of a Universal Language?2 people · 2 mentions · first linked 2026-03-24
What Hacker News says?
Seems conceptually connected to the "repeat yourself" hack that improves models by duplicating layers: https://dnhkng.github.io/posts/rys/
Comment linking /posts/rys/
on “How concerned should we be about Astra's recurrent architecture?”Submitted I got an Nvidia GH200 server for €7.5k on Reddit and converted it to a desktop as a story · 373 points
HN discussionThe recurrent depth sounds a lot more like what is described in https://dnhkng.github.io/posts/rys/ - a way to add depth to a network without increasing the number of parameters. e: While the actual CoT in neuralese paper is Facebook's Coconut https://arxiv.org/abs/2412.06769 - not sure if any production models use that one.
Comment linking /posts/rys/
on “What is nueralese and why is it bad”Submitted Show HN: I built a hardware quantum RNG and wired it into a Magic 8-Ball as a story · 7 points
HN discussionConceptually, this seems a good direction. The other piece that has always struck me as a huge inefficiency with current usage of LLMs is the hoops they have to jump through to make sense of existing file formats - especially making sense of (or writing) complicated semi-proprietary formats like PDF, DOC(X), PPT(X), etc. Long-term prediction: for text, we'll move away from these formats and towards alternatives that…
Comment linking /posts/rys/
on “Kotlin creator's new language: a formal way to talk to LLMs instead of English”Submitted Show HN: How I Topped the HuggingFace Open LLM Leaderboard on Two Gaming GPUs as a story · 7 points
HN discussionYes! I tried that pretty early on, the its basically never good. Its described in the the section: https://dnhkng.github.io/posts/rys/#the-beginning-of-llm-neu...
Comment linking /posts/rys/
on “Show HN: How I Topped the HuggingFace Open LLM Leaderboard on Two Gaming GPUs”Submitted LLM Neuroanatomy II: Modern LLM Hacking and Hints of a Universal Language? as a story · 2 points
HN discussionI keep thinking of the RYS (Repeat Yourself) experiment of simply looping some of the inner layers of LLMs for better results and wonder if any progress was made on it. https://dnhkng.github.io/posts/rys/ Feels it should be straightforward to integrate in LLMs a network to control the looping. Or just duplicate entire blocks of layers after the initial training.
Comment linking /posts/rys/
on “Is One Layer Enough? A Single Transformer Layer Matches Full-Parameter RL Train”Submitted LLM Neuroanatomy II: Modern LLM Hacking and Hints of a Universal Language? as a story · 2 points
HN discussionYes you can! Well, mostly, depends on how pedantic you are with definitions: you can transplant layers but not weights, which in common parlance are conceptually similar. But usually it isn’t a good idea for a few reasons. There’s a really fascinating example[1] where a guy identifies a particular set of layers and transplants them. Overgeneralizing, early layers are encoders and the later layers are decoders and in…
Comment linking /posts/rys/
on “Anthropic Accuses Alibaba of 'Illicitly' Accessing AI Models”Submitted Show HN: I built a hardware quantum RNG and wired it into a Magic 8-Ball as a story · 1 points
HN discussionThis reminds me of https://dnhkng.github.io/posts/rys/ David looks into the LLM finds the thinking layers and cut duplicates then and put them back to back. This increases the LLM scores with basically no over head. Very interesting read.
Comment linking /posts/rys/
on “Which one is more important: more parameters or more computation? (2021)”Submitted Do LLMs Break the Sapir-Whorf Hypothesis? as a story · 1 points
HN discussionIf I understand it correctly, this is based on the "RYS" architecture (or findings) by David Ng? ( https://dnhkng.github.io/posts/rys/ ) And, related: if there are small subsets of layers that can be looped inside LLMs to improve their reasoning, and if the layers to loop change depending on the competencies used by the LLM in that particular context, has anyone yet tried to build and train an LLM that can decide wh…
Comment linking /posts/rys/
on “OpenMythos: A looped transformer take on how Claude Mythos might work”Submitted The Silicon Leash – Why ASI Needs Human Cooperation as a story · 1 points
HN discussionapparently you can straight up duplicate/add/rearrange layers without changing any of the weights and get better results as well - https://dnhkng.github.io/posts/rys/
Comment linking /posts/rys/
on “Apple: Embarrassingly Simple Self-Distillation Improves Code Generation”Source for those interested https://dnhkng.github.io/posts/rys/
Comment linking /posts/rys/
on “A global workspace in language models”Found it: https://news.ycombinator.com/item?id=47500709 Part 3 might be the best introduction: https://dnhkng.github.io/posts/sapir-whorf/ tl;dr: Based on experiments with similar prompts translated to different languages LLM layers group into three phases: the first decodes from the source language into an abstract space, the middle does something, then there's a last part where the abstract result gets transformed…
Comment linking /posts/sapir-whorf/
on “A global workspace in language models”It's been shown that LLMs use their outer layers to decode from and encode to language, while their middle layers deal in language-independent abstract concepts. This means that the same question or statement in different languages activates the outer layers differently but produces the same patterns in the middle layers. Check this article with cool visualizations (btw, this is one of the articles mentioned also by…
Comment linking /posts/sapir-whorf/
on “A global workspace in language models”There is an interesting on-going research https://dnhkng.github.io/posts/sapir-whorf/ that shows LLMs think in a language-agnostic way. (It will probably get posted to HN after it is finished.)
Comment linking /posts/sapir-whorf/
on “Writing Lisp Is AI Resistant and I'm Sad”Would recommend reading https://dnhkng.github.io/posts/rys/ and check his github code to reproduce the findings. Easy to way to start hacking LLMs; there's much of value there and a fun way to get into it before tackling heavy math / CS topics.
Comment linking /posts/rys/
on “Ask HN: How to Break into AI Engineering (Revisited)”You can also edit binary distributions of models with means besides changing their weights. See "LLM Neuroanatomy: How I Topped the LLM Leaderboard Without Changing a Single Weight." On the spectrum of: careful engineering--hacking--mad science This kind of thing falls far towards the mad science end of the scale, but has proven effective. https://dnhkng.github.io/posts/rys/
Comment linking /posts/rys/
on “There is minimal downside to switching to open models”You're right that my quote was misleading, I overlooked "the weird part" in the post because it didn't seem new to me either. Here's the section in the original post that covers it: https://dnhkng.github.io/posts/rys/#the-brain-scanner All heatmaps are split by tasks and show an optimal point for each. The resulting routing he chose is a trade-off for both tasks, there isn't much else to do unless you intend to trai…
Comment linking /posts/rys/
on “Show HN: Duplicate 3 layers in a 24B LLM, logical deduction .22→.76. No training”
