Interviewing Ross Taylor on LLM reasoning, Llama fine-tuning, Galactica, agents

Startup Journey

Ross Taylor recounts the early days of his startup journey, highlighting the unexpected success of Papers with Code. He and his co-founder, Rob, came from diverse backgrounds but resonated on similar problems within the ML space 1. Their initial experiments led to the creation of a website to solve the problem of finding code for research papers, which surprisingly took off 1. This success eventually shifted their focus towards research and the development of large language models like Galactica 2.

We hacked around with this little website to solve this really basic problem. Like, ok, I'm trying to reproduce this paper, but I can't find the code. And that was the thing that really blew up to our expectation.

--- Ross Taylor

Their journey underscores the importance of adaptability and seizing unexpected opportunities in the tech world.

   

Papers with Code

Papers with Code started as a simple solution to find code for research papers but evolved into a comprehensive platform indexing various ML artifacts. Ross Taylor explains that their users' needs drove this expansion, leading to the creation of leaderboards and indexing state-of-the-art methods for different benchmarks 3. The platform's organic growth and collaboration with the PyTorch community eventually led to its acquisition by Meta 3.

We kind of expanded away from that. Just a very simple use case of finding code to indexing various artifacts in ML.

--- Ross Taylor

This evolution highlights the platform's significant impact on the ML community and its role in promoting open-source research.