Top all-time · #16055
Ofir Press
Personal blog · ofir.io · on HN as ofirpressai
Ofir Press's personal blog at ofir.io covers ai. Since 2016, 4 different Hacker News readers have linked it 5 times. It is still writing: the last post was on 2026-10-13, and you can follow it by feed.
Too few sampled comments (1) to say how often HN recommends it.?
Why HN links it
HN readers link Ofir Press's post on the use case for relative position embeddings in transformers. It is cited as a reference in a discussion comparing RoPE and ALiBi for extending context length.
Summarised by Claude from a sample of 1 HN comments that link here. Each comment is marked recommending, critical or neutral (citing without judging); how this works.
My thoughts on deep learning for NLP
What Hacker News says?
Why use RoPE over Alibi? Truly relative positional embeddings help increase context length even after training. Ref: [1]: https://ofir.io/The-Use-Case-for-Relative-Position-Embedding...
Comment linking /The-Use-Case-for-Relative-Position-Embeddings
on “Llama 2 – Meta AI”Submitted How to Start Learning Deep Learning as a story · 264 points
HN discussionSubmitted An Introduction to Deep Learning for Language Modeling as a story · 206 points
HN discussionSubmitted Neural Language Modeling from Scratch as a story · 97 points
HN discussionSubmitted Shortformer: Better Language Modeling using Shorter Inputs [pdf] as a story · 64 points
HN discussion