Lumen

Lumen

Eager to know more, about the world, about the intelligence, and about myself.
github
HomeThoughtsLearningArchives

Assorted links 0, 5.8.23

#assorted-links#LLM384
AI-generated summary
This is a new series of posts where the author shares interesting articles, projects, and papers they discover through their RSS reading. They are grateful to RssHub and Miniflux for enabling them to read captivating content freely. They are inspired by bloggers like Tyler Cowen and Simon Willison and want to share their personal interests in a similar manner. The author also includes amusing links in their collection. They hope their posts can provide interesting updates to more people and welcome discussions on various subjects. They mention some specific links they found interesting, such as a causal language model for coding, a scheme for inferring reward from interaction, a leaked Google document on the development of LLM, SparseGPT for pruning weight in a language model, a post on RLHF, job interview prompts, and an introduction to Kalman filter. They also mention some "weird stuff" without further explanation.

About assorted links, why I'm writing this?#

This is a new series of posts. In this series, I will share interesting articles, projects, and papers that I discover through my RSS reading. I am grateful to RssHub by DIYgod and Miniflux for enabling me to read captivating content from various sources across the internet freely and effectively.

Tyler Cowen frequently shares intriguing links with brief summaries on Marginal Revolution. I also gain valuable information from Simon Willison, who shares interesting news in the field of LLM.

Inspired by these bloggers, I want to share my personal interests in a similar manner. These posts were initially my reading notes, but sharing them serves as an effective way to organize the content. Besides valuable information, my collection of links includes also stuffs I find amusing.

It would be great if my post can make some interesting updates available to more people. I also welcome discussions on various subjects, as I believe they are mutually beneficial for all of us.

Miniflux Reading#

  1. replit-code is a causal language model that is specialized on coding. I'm actually more interested at the causal part of the LLM.

  2. Microsoft has designed a scheme that infers reward from interaction. Check it out if you can use it as the training bit for the real world interaction

  3. The leaked Google document on analysing current development of LLM is highly recommended to read. Bascially Google believes that the open source LLM is having rapid development and has several advantages, which close sourced LLM from Google or OpenAI cannot offer. I'm personally excited that the open-source community has shown great power in the competition on LLM, this could eventually bring back the openness in the research field as well as give us more control over how LLM would affect our lifes.

  4. SparseGPT can prune up to 60% of the weight of a LLM without any retraining.

  5. A post from Chip Huyen that discuss more about the RLHF. Chip Huyen is a blog writer that has many valuable articles, which I don’t have time to read through. She has also shared her booklist in 2022.

  6. Job interview prompts shared from Alexander Young. He also has some blog posts that shares how to write essay with the help of ChatGPT and also study.

  7. A post from Reality bytes to introduce Kalman filter for you.

Some weird stuff …

image

Loading...
Ownership of this post data is guaranteed by blockchain and smart contracts to the creator alone.