Title

Speaker(s): Aparna Dhinakaran, Amber Roberts

Video URL: https://www.youtube.com/watch?v=K6Wui3mn-uI

Summary: In this talk, the speakers emphasize the importance of staying current with rapid advancements in AI research. They provide guidance on key resources, outline different types of AI papers, and conduct a live demonstration on how to analyze a research paper. Additionally, they highlight the value of actively engaging with the research community and contributing to it yourself.

Timestamps

• 00:00 - Introduction to the session and speakers
• 04:30 - Overview of the talk
• 05:50 - Reasons to read research papers
• 08:50 - Resources on how to get started
• 11:10 - Types of AI research papers
• 35:40 - How to go through a paper, example: "Mixtral of Experts"
• 01:00:00 - "What's next?" - Tips on how to get started
• 01:02:30 - Q&A

Key Takeaways

• You don't necessarily need to search through academic databases, there are other spaces. For example: a lot of discussions around papers and hints about new interesting ones happen on the social media platform X (former Twitter).
• There are different types of AI papers, the ones mentioned are: survey papers, benchmark papers, and breakthrough papers - each serving a different purpose in academic and practical contexts.
• Survey Papers provide overviews of specific topics, consolidating existing research over a certain time period, e.g. 1,5 year.
• Conceptual Papers introduce new theories or methodologies that can shape future research.
• Benchmark Papers establish new datasets or evaluation standards for comparing AI models.
• Application Papers demonstrate innovative uses of AI in various fields, showing practical implementations.
• Experimental Papers focus on experiments and results, offering insights through empirical studies.

Questions/Discussion Points

• Very benenficial divison of different types of papers and how to spot them, but there are (IMO) better examples of demonstrating how to approach reading research papers (in general, not particularly in AI/CS fields), .e.g: How To Read A Paper Quickly & Effectively With AI in 08:20 minutes, How to read a research paper | Luís and João Batalha and Lex Fridman in 06:16 minutes, Become an AI Expert by Reading Like a Pro in 08:16 minutes.

Links/Resources

• A Survey of Large Language Models
• Retrieval-Augmented Generation for Large Language Models: A Survey
• HellaSwag: Can a Machine Really Finish Your Sentence
• Mistral AI Mixture of Experts