I've been waiting for an update to this study for a long time. It confirms in AI a trend happening across science & tech fields: China is gaining share in elite research, and the US is beginning to lose attractiveness among top talent. A few relevant stories ed below.
RESEARCH
-
MICS Launches CroissantLLM, Bilingual Open Source French-English Model
By
–
[#Article] Generative AI: MICS Launches CroissantLLM, a Bilingual Open Source French-English Model https://actuia.com/actualite/ia-generative-le-mics-lance-croissantllm-un-modele-open-source-bilingue-francais-anglais/
… #AI #ArtificialIntelligence #GenerativeAI -
Anthropic Presents Claude 3, Next Generation AI Models
By
–
[#Article] Generative AI: Anthropic Unveils 3rd Generation of Claude Model Family https://actuia.com/actualite/ia-generative-anthropic-devoile-la-3eme-generation-de-sa-famille-de-modeles-claude/
… #AI #artificialintelligence #GenAI -
Llama 7B Training on Single RTX 4090 GPU Memory Efficient
By
–
For the first time, we show that the Llama 7B LLM can be trained on a single consumer-grade GPU (RTX 4090) with only 24GB memory. This represents more than 82.5% reduction in memory for storing optimizer states during training. Training LLMs from scratch currently requires huge
-
ChatGPT’s Physical Limitations: Why AI Can’t Paint Brushes
By
–
I mean it's not lying… ChatGPT will likely never paint with a paintbrush…
-
Self-Retrieval: Building Information Retrieval System with One LLM
By
–
Self-Retrieval: Building an Information Retrieval System with One Large Language Model Tang et al.: https://
arxiv.org/abs/2403.00801 #ArtificialIntelligence #DeepLearning #MachineLearning -
NEJM Review Table of Healthcare AI Interventions
By
–
In the new @NEJM review I cited there is a Table of things that can be done
-
Three Landmark AI Models: GPT-3, Gato, and Flamingo
By
–
Only 3 times before: GPT3, Gato and Flamingo and I coded my first neural net exactly 30 years ago
-
AI Development Approach Grounded in Local Perspectives
By
–
Stanford HAI & The @Asia_Foundation call for an approach to AI in global development that is grounded in local perspectives and tailored to the specific circumstances of Global Majority countries. Read their joint response to @USAID
’s request for comment: https://
stanford.io/3wQ2JFe -
Groupthink and consensus in complex AI questions
By
–
"We know/don't know" is an extremely normie way of referring to groupthink. Many complex questions don't have consensus answers.