@karpathy 1/ I have been working on a similar concept for my company using Claude Managed Agents. We have several specialized agents running remotely on the Claude platform that employees can interact with directly in Slack. The agents have access to several of our corpo knowledge systems
@jerryjliu0 2/ RAG presentations never stress out the importance of the retrieval part. If the search is not relevant, you will get a crappy answer from the LLM: "junk-in, junk-out" syndrome. So what we need is RRAG: Relevant Retrieval Augmented Generation
See https://t.co/cN8OrFfbtb
HTH!
@jerryjliu0 Nice presentation Jerry, thanks! Two comments:
1/ RAG presentations always explain in great details the pipeline (slide 10 of your presentation) but never WHY we have to go through all this trouble. The reason is simple, though: the number of tokens the LLM can ingest is limited
@garnulfthegrey@ylecun@ClementDelangue@DrJimFan@emmanuel_2m +1. Actually, LangChain and LlamaIndex are supposed to be such frameworks, but they are very far from the quality of what MacApp was in terms of documentation, code samples, readable code, ease of use, learning curve, clarity of the classes and methods, etc. I suffer every day…
I wrote a little guide at Asmodee to help my colleagues #remoteworking during the #coronavirus episode, and I was allowed to make it public. Hopefully this can help other people, feel free to share. Enjoy and stay safe!
https://t.co/ik9pG5FwAp
Hey @RERB , j’ai une idée absolument géniale : et si on activait les climatisations en cette période de canicule ? Depuis le début de la semaine, aucune rame « modernisée » que j’ai prise n’avait de clim opérationnelle...