Top all-time · #11015
Ken Schutte
Personal blog · kenschutte.com · on HN as ks2048programmingai
Ken Schutte's personal blog at kenschutte.com covers programming and ai. Since 2023, 7 different Hacker News readers have linked it 20 times, most often ““Gzip beats BERT?” Part 2: dataset issues, improved speed, and results”. It is still writing: the last post was on 2026-03-14, and you can follow it by feed.
Too few sampled comments (6) to say how often HN recommends it.?
Why HN links it
HN readers link Ken Schutte's posts as rebuttals to a 2023 ACL paper claiming gzip with kNN rivals BERT, arguing the result depended on test-set leakage and an unfair top-2 versus top-1 comparison.
Summarised by Claude from a sample of 6 HN comments that link here. Each comment is marked recommending, critical or neutral (citing without judging); how this works.
Most-linked posts?
- “Gzip beats BERT?” Part 2: dataset issues, improved speed, and results4 people · 5 mentions · first linked 2023-07-29
- Bad numbers in the “gzip beats BERT” paper?4 people · 4 mentions · first linked 2023-07-17
What Hacker News says?
Unfortunately GZIP won't beat LLMs for text classification. The research you cited is just poorly done science that has been widely debunked. The original paper compared top-2 accuracy of GZIP with top-1 accuracy with BERT. The dataset also contains a lot of train/test data leakage. See this article for the rebuttal: https://kenschutte.com/gzip-knn-paper/ and this thread for a previous discussion on hackernews: http…
Comment linking /gzip-knn-paper/
on “NY Times copyright suit wants OpenAI to delete all GPT instances”Submitted Bad numbers in the “gzip beats BERT” paper? as a story · 41 points
HN discussionGetting (back) into development for Apple TV. Released an updated version of free app [0] for watching the news show Democracy Now [1]. Let's you browse 29 years of back episodes. Learning SwiftUI - mostly great, but when there's problems, it can be pretty frustrating. [0] https://kenschutte.com/democracy-now/ [1] https://www.democracynow.org/
Comment linking /democracy-now/
on “Ask HN: What are you working on? (February 2025)”Submitted Bus travel from Lima to Rio de Janeiro as a story · 2 points
HN discussionNot LLM, just BERT, also did not actually outperform it. source: https://kenschutte.com/gzip-knn-paper/
Comment linking /gzip-knn-paper/
on “Ts_zip: Text Compression Using Large Language Models”Submitted Swap_8_and_9: A simple import can modify the Python interpreter as a story · 2 points
HN discussionI recently used popcount for an audio-fingerprinting project [0] to compute the difference in binary "audio fingerprints". With AVX2 and 4 cores, my desktop managed 448 billion bit-diff calcs per second! [0] https://kenschutte.com/phingerprint/
Comment linking /phingerprint/
on “You Won’t Believe This One Weird CPU Instruction (2019)”Submitted “Gzip beats BERT?” Part 2: dataset issues, improved speed, and results as a story · 2 points
HN discussionDid a trip like this a few years ago. Highly recommended, if you have the opportunity! Wanted to do a write-up like this, but only got as far as the map, https://kenschutte.com/europe-2023/ I wonder if they took the new "Rail Baltica" through the Baltics? That was one of weaker links in the train route - I used a bus between Vilnius and Riga.
Comment linking /europe-2023/
on “Interrail: 6,379Km and 13 Countries over 7 weeks”Submitted Did Musk steal his new X logo? (Glyph analysis in Python) as a story · 1 points
HN discussionI was in Brazil for the first time last year and was very impressed with the trees. Two examples right from downtown São Paulo, https://kenschutte.com/lima-to-rio-by-bus/images/trees.jpg
Comment linking /lima-to-rio-by-bus/images/trees.jpg
on “Some Unusual Trees”People interested in Forth might want to also check out PostScript (similar and stack-based). I recently wrote a blog post with a very brief PostScript tutorial and an example program translated line-by-line to Python https://kenschutte.com/postscript-vs-python/
Comment linking /postscript-vs-python/
on “My history with Forth, and stack machines (2010)”haven’t looked through this post yet but I think this is what you have in mind: https://kenschutte.com/gzip-knn-paper/
Comment linking /gzip-knn-paper/
on “Show HN: 78% MNIST accuracy using GZIP in under 10 lines of code”Well…one that peeks at the test set labels. https://kenschutte.com/gzip-knn-paper2/
Comment linking /gzip-knn-paper2/
on “Decoding the ACL Paper: Gzip and KNN Rival Bert in Text Classification”Further analysis shows that it doesn’t perform well at all—successes are tied to things like test set leakage. https://kenschutte.com/gzip-knn-paper2/ This paper isn’t any surprisingly effective result. It’s thoroughly shoddy scholarship by which the authors should feel embarrassed.
Comment linking /gzip-knn-paper2/
on “Decoding the ACL Paper: Gzip and KNN Rival Bert in Text Classification”Chromaprint computes “features” roughly 8 times per second. You can do a brute-force search checking different times in a song or potentially do some more fancy indexing once you have the features. (I did some experiments with Chromaprint - described here, https://kenschutte.com/phingerprint/ )
Comment linking /phingerprint/
on “How does Shazam work? (2022)”This looks like a nice rundown of how to do this with Python's zstd module. But, I'm skeptical of using compressors directly for ML/AI/etc. (yes, compression and intelligence are very closely related, but practical compressors and practical classifiers have different goals and different practical constraints). Back in 2023, I wrote two blog-posts [0,1] that refused the results in the 2023 paper referenced here (bad…
Comment linking /gzip-knn-paper2/
on “Text classification with Python 3.14's ZSTD module”This took me an unreasonable amount of time to find, but here it is https://kenschutte.com/gzip-knn-paper2/ The moral: the methodology is cool, but implementation details matter, i guess...
Comment linking /gzip-knn-paper2/
on “Mutually Assured Recursion”I just wrote a blog post [0] on a project that I did with audio-fingerprinting and to my surprise, my desktop could do 369 billion bit-comparisons per second (popcnt using AVX2 instructions, 4 simultaneous processes). [0] https://kenschutte.com/phingerprint/
Comment linking /phingerprint/
on “Do You Know How Much Your Computer Can Do in a Second?”
