From 4d9b70fbab5b0b4d0742e08b98d05bf0633a8b87 Mon Sep 17 00:00:00 2001 From: Omar Khattab Date: Wed, 4 Sep 2024 10:35:51 -0700 Subject: [PATCH] Update 2024.09.impact.md --- 2024.09.impact.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/2024.09.impact.md b/2024.09.impact.md index f57c7f5..40ecf2c 100644 --- a/2024.09.impact.md +++ b/2024.09.impact.md @@ -97,4 +97,4 @@ When you read the previous guideline (#5), it's very natural to wonder: Where mi The answer in practice is that it's possible for most of your time OSS time to be spend on doing new, exciting research. In fact, a major advantage of the style of research in this guide is that it creates recognizably important problems in which you have a very large competitive advantage. This is true in terms of the intuitive recognition of problems extremely early, having a far more instinctive understanding of the problem than others, having direct feedback on prototypes of your approaches, having access to wonderful collaborators who understand the significance of the problems, and also having great "distribution channels" that ensure that every new paper you do in this space will have a receptive audience and will reinforce your existing platform. -Just to illustrate, ColBERT isn't one paper from early 2020. It's probably around ten papers now, with investements into improved training, lower memory footprint, faster retrieval infrastructure, better domain adaptation, and better alignment with downstream NLP tasks. Similarly, DSPy isn't one paper but is a large collection of papers on programming abstractions, prompt optimization, and downstream programs. So many of these papers are written different, amazing primary authors. A good open-source artifact creates modular pieces that can be explored, owned, and grown by new researchers and contributors. +Just to illustrate, ColBERT isn't one paper from early 2020. It's probably around ten papers now, with investements into improved training, lower memory footprint, faster retrieval infrastructure, better domain adaptation, and better alignment with downstream NLP tasks. Similarly, DSPy isn't one paper but is a large collection of papers on programming abstractions, prompt optimization, and downstream programs. So many of these papers are written by [different, amazing primary authors](https://github.com/stanfordnlp/dspy?tab=readme-ov-file#dspy-programmingnot-promptingfoundation-models). A good open-source artifact creates modular pieces that can be explored, owned, and grown by new researchers and contributors.