Hacker News new | past | comments | ask | show | jobs | submit login

There was a series of articles on lossless and lossy and compression techniques in, I think, PC Magazine that I read as a kid and made a big impression on me. I didn't, like, end up ready to write code to build Huffman trees or anything, but it did change it from a total black box into a bunch of smaller pieces each of which a mere mortal can understand.

The compression rabbit hole can be a rewarding one to go down if you haven't. Besides the tech itself making a lot of things work better (like a Web built on human-readable formats, or more specialized stuff like Parquet or other columnar formats, or video on the lossy side), it can give you some tools or perspective that apply elsewhere, e.g. to other probablistic stuff like caching and predictions for lossless compression, or other signal-processing stuff for lossy compression.




Another interesting facet is that, according to some schools of thought, compression and AI are equivalent problems: The better you understand something, the less you need to memorize to reproduce it.

https://en.wikipedia.org/wiki/Hutter_Prize (and https://en.wikipedia.org/wiki/AIXI)

Of course, large language models are (by definition) currently going the other direction, but it remains to be seen whether that leads to artificial intelligence (whatever that ends up meaning).


> Of course, large language models are (by definition) currently going the other direction ...

How so? Aren't the networks' weights orders of magnitude smaller than the training data?


I interpreted that statement as saying the current practice is to make LLMs larger and larger (so they effectively memorize more and more data) to make them more powerful, but from the perspective of information theory, if models were powerful and "understanding", then models could stay the same size and become more and more powerful as they get increasingly better at compressing the available information. I am not sure if this interpretation was what was meant though.


I believe the parent poster's point is: LLMs are more effective when they use more memory, meaning the less they are forced to compress the training data, the better they perform.


But they don't losslessly recreate the training data.


Maybe a naive question: are LLMs really going the other way? My intuition is that the model weights are much smaller than the information encoded in them.


Once ChatGPT was trained on "the internet of 2021", how big is the remaining model? Is ChatGPT effectively a compressed version of the internet?


Was it maybe Dr Dobb's? Mark Nelson had an excellent series of articles about compression which opened up that world for me. I ended up working on compression for my PhD many years later.

I can't find any archives for Dr Dobb's, but some articles are in his personal site, for example this one about arithmetic coding: https://marknelson.us/posts/1991/02/01/arithmetic-coding-sta...


It looks like the ones I were PC Magazine, https://books.google.com/books?id=gCfzPMoPJWgC&lpg=PP1&pg=PA... and https://books.google.com/books?id=eX8w8B-OhIIC&lpg=PA371&pg=... .

Much appreciation to all the folks who put in the work to make these things interesting and accessible.


Hi I am a neophyte interested in compression. It's difficult to find communities regarding compression online. I am looking for a guide. Is there any place that I can dm you?


I found their Encyclopedia page[0], but I suspect you meant something better than this?

Any further recollections to help drill down and find it?

- [0]: https://www.pcmag.com/encyclopedia/term/data-compression


It looks like it was Barry Simon in "Lab Notes" in PC Magazine of June and July 1993:

https://books.google.com/books?id=gCfzPMoPJWgC&lpg=PP1&pg=PA...

https://books.google.com/books?id=eX8w8B-OhIIC&lpg=PA371&pg=...




Consider applying for YC's Summer 2025 batch! Applications are open till May 13

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: