Yangfeng Ji

5 Followers
30 Following
44 Posts
Assistant Professor @ UVa CS 
#NLProc #ML #Coffee
Personal webpagehttps://yangfengji.net/
Group webpagehttps://uvanlp.org/
Mediumhttps://medium.com/@yangfengji

Perspective paper by myself and David Krakauer is now published in PNAS:

"The Debate Over Understanding in AI's Large Language Models".

PNAS: https://www.pnas.org/doi/10.1073/pnas.2215907120

arXiv: https://arxiv.org/abs/2210.13966

I love #obsidian
Tweet / Twitter

Twitter

RT @yangfeng

Our group released a Python package of data valuation in machine learning, Valda. It supports five methods (LOO, Influence Function, TMC-Shapley, Beta-Shapley, and CS-Shapley) via a unified API. Please try it out if you are interested:
uvanlp.org/valda/

https://twitter.com/yangfeng_ji/status/1619365867648843777?s=46&t=LZRXyXdZET5khWrVCzE_6A

Yangfeng Ji on Twitter

“Our group released a Python package of data valuation in machine learning, Valda. It supports five methods (LOO, Influence Function, TMC-Shapley, Beta-Shapley, and CS-Shapley) via a unified API. Please try it out if you are interested: https://t.co/IrkzJqbMlr @stephschoch”

Twitter

It's amazing how #ChatGPT evolves every day.

When I asked "can I build a copy of you?" Here are what I got:

Left: no answer yesterday
Right: a very good answer today, although I found it a little intimidating 😅

Not sure anyone agrees with me, but #ChatGPT is not a research demo or prototype on LLMs; it’s a product. We all know the PageRank algorithm, but have we ever asked Google to write a paper about how its search engine exactly works?
Meanwhile, it doesn’t make sense to think the success of a product is just because of one algorithm or model.
(Cross-posting from Twitter)

Please check out the welcome page: https://neurips.cc/virtual/2022/index.html

There is a lot going on.
Keep this map of the venue handy!

Thanks for all the responses! It may help to distinguish two types of speculation (cf. https://terrytao.wordpress.com/career-advice/theres-more-to-mathematics-than-rigour-and-proofs/): "pre-rigorous speculation", in which one asks "dumb" questions (cf. https://terrytao.wordpress.com/career-advice/ask-yourself-dumb-questions-and-answer-them/) before fully knowing the field, and "post-rigorous speculation", in which one shares informed insights and opinions from one's rigorous understanding of the field.

I think we should encourage both types, in appropriate venues of course (and separated from traditional "rigorous" work).

There’s more to mathematics than rigour and proofs

The history of every major galactic civilization tends to pass through three distinct and recognizable phases, those of Survival, Inquiry and Sophistication, otherwise known as the How, Why, and Wh…

What's new

I didn’t say the Cicero paper should be rejected and I would not necessarily advocate for that.

But there is something that makes me grumpy. Science, and, more notably, Nature have become the place for big-flashy-things. Often (but not always) these big flashy things have a moderate scientific innovation and a huge engineering achievement. That’s fine. Except Science and Nature are received by a wider, more general audience that is less able to parse the nuance. This in turn drives hype cycles.

One of the cats always joins me, whenever I sit down with my laptop