We picked 50 paper/models/blogs across 10 fields in AI Eng: LLMs, Benchmarks, Prompting, RAG, Agents, CodeGen, Vision, Voice, Diffusion, Finetuning. If you're starting from scratch, start here.
Thanks for creating this comprehensive list - this is gonna be a long read ;)
For other readers like me that sometimes struggle (or are willing to help others - @authors) to understand a paper, please consider joining the community I’ve created on Discuna where we can share questions and other comments directly in these papers *within a community*:
- currently just me, but I’ll do my best to respond! I've already added the papers from sections 1-5 to this community.
Discuna (discuna.com) is a free non-profit tool I programmed that is like Discord but in addition to text channels you also have PDF channels that allow for sharing comments directly inside PDF documents within communities. (In principle you could also program your own channels!)
PS I hope I did not violate some rules with this post but I genuinely believe this tool will help the community.
Great stuff! Uploaded all to Zotero and will get to reading. Unfortunately, the Discuna invite link seems to be expired. Any news when you'll update it?
Thank you to the authors for their very patient and meticulous organization of so much learning material. This has long exceeded the content that should be on your reading list. It covers GitHub, YouTube, podcasts, official websites, and more. I feel your dedication and seriousness, and I have also compiled a link that is friendly to both Chinese speakers and non-native English speakers. The paper section has all been changed to links that can be used with immersive translation plugins. If the authors see this and feel it is inappropriate, please let me know, and I will delete the comment and close the link's open access immediately. Once again, thank you for the authors' heartfelt https://t0o1vpznu8.feishu.cn/docx/WS6OdpgMoobYq5x66KwcsSEInIQ?from=from_copylink
yes :) because its more about "families" of papers than the individual paper. This may cause uneven workloads, but also reflects the fact that older papers (GPT1, 2, 3) are less relevant now that 4/4o/o1 exist, so you should proportionately spend less time each per paper, and kind of lump them together and treat them as "one paper worth of work", simply because they are old now and have faded to rough background knowledge that you'll roughly be expected to have as an industry participant
I checked it out, the Discord content built by Kristin at https://app.discuna.com/invite/ai_engineer. It selects 5 articles from each section, including not only papers but also podcasts, from all the ones that already have links. So all the links listed actually belong to the reading list, and the ones at the end of each paragraph are carefully updated readable papers after the technical updates.
Thanks for creating this comprehensive list - this is gonna be a long read ;)
For other readers like me that sometimes struggle (or are willing to help others - @authors) to understand a paper, please consider joining the community I’ve created on Discuna where we can share questions and other comments directly in these papers *within a community*:
https://app.discuna.com/invite/ai_engineer
- currently just me, but I’ll do my best to respond! I've already added the papers from sections 1-5 to this community.
Discuna (discuna.com) is a free non-profit tool I programmed that is like Discord but in addition to text channels you also have PDF channels that allow for sharing comments directly inside PDF documents within communities. (In principle you could also program your own channels!)
PS I hope I did not violate some rules with this post but I genuinely believe this tool will help the community.
awesome! all the best! Will edit to include your invite link.
Great, thanks so much! Added sections 6-10 as well^^
Link expired?
hi, thanks for pointing this out! The link should work now again (accidentally set expiration date of invitation link to 1 week)
This is definitely ideal way to learn, do you have any recommendations for someone doing it alone.
start from 1 and go to 50 :)
Thank you, this is going to be a great year. Happy New Year
The “2024 SOTA Agents designs at Neurips” link is broken
https://www.latent.space/p/2024-agents
Thank you for putting together and sharing this list! ☺️
Great stuff! Uploaded all to Zotero and will get to reading. Unfortunately, the Discuna invite link seems to be expired. Any news when you'll update it?
Thank you to the authors for their very patient and meticulous organization of so much learning material. This has long exceeded the content that should be on your reading list. It covers GitHub, YouTube, podcasts, official websites, and more. I feel your dedication and seriousness, and I have also compiled a link that is friendly to both Chinese speakers and non-native English speakers. The paper section has all been changed to links that can be used with immersive translation plugins. If the authors see this and feel it is inappropriate, please let me know, and I will delete the comment and close the link's open access immediately. Once again, thank you for the authors' heartfelt https://t0o1vpznu8.feishu.cn/docx/WS6OdpgMoobYq5x66KwcsSEInIQ?from=from_copylink
I'm sorry, maybe I'm not that smart, is it supposed to be reading the papers in the link, and if so, should there be more than 50, isn't it?
Agreed, it says 5 papers per section, but there are >5 in most. I'm a little confused as well.
yes :) because its more about "families" of papers than the individual paper. This may cause uneven workloads, but also reflects the fact that older papers (GPT1, 2, 3) are less relevant now that 4/4o/o1 exist, so you should proportionately spend less time each per paper, and kind of lump them together and treat them as "one paper worth of work", simply because they are old now and have faded to rough background knowledge that you'll roughly be expected to have as an industry participant
Awesome, makes sense. Thank you!!
I checked it out, the Discord content built by Kristin at https://app.discuna.com/invite/ai_engineer. It selects 5 articles from each section, including not only papers but also podcasts, from all the ones that already have links. So all the links listed actually belong to the reading list, and the ones at the end of each paragraph are carefully updated readable papers after the technical updates.