LLMS

→ View original post on X — @jeffdean

Google AI Research 2022: Language Models and Generative AI

By

@jeffdean

–

18 January 2023 20h39

Excited to share the first of a series of @GoogleAI blog posts summarizing our research work from 2022. This covers language & multimodal models, computer vision, and generative models. We'll have ~7 posts covering other areas over next few weeks! https://
ai.googleblog.com/2023/01/google
-research-2022-beyond-language.html
…

Improved Documentation for Custom LangChain Agents

By

–

18 January 2023 18h54

Better documentation for understanding how custom agents work @samcwl added some much needed comments in both the documentation and the below notebook on the expected LLM output to get it to work with custom agents https://
langchain.readthedocs.io/en/latest/modu
les/agents/examples/custom_agent.html
…

AI21Labs Experimental Models Now Available for Integration

By

–

18 January 2023 18h54

Also, highlighting one of the earlier points, its now possible to use some of the @AI21Labs experimental models! “`llm = AI21(maxTokens=10, model="j1-grande-instruct")“` Looking forward to playing around with this

v0.0.65 Release: Documentation Fixes and AI Model Support

By

–

18 January 2023 18h54

v0.0.65 Doc Fixes – Francis, Chetanya Rastogi
Prompt improvements – @fpingham @brunotorious Support for @AI21Labs experimental models – xettrisomeman
Fix default @CohereAI embedding model – Sasmitha Manathunga Some doc improvements, listed

FAISS Vectorstore Saving and Loading Guide

By

–

18 January 2023 18h54

Instructions for saving and loading FAISS vectorstore You just want to create your vectorstore once, and then pass it around. We added some easy documentation on how to do so. https://
langchain.readthedocs.io/en/latest/modu
les/utils/combine_docs_examples/vectorstores.html#saving-and-loading
…

→ View original post on X — @goodside

OpenAI vs Anthropic: human ratings vs AI consistency judgment

By

@goodside

–

18 January 2023 17h28

It doesn’t translate principles into code. Oversimplifying, the difference is that OpenAI’s models tune on generations rated as “perfect” by humans, or by RL imitating humans, and Anthropic’s tune on those judged by an AI to be consistent with a human-written “constitution”.