On his world tour @sama dialed into a Beijing conference today and called for US-China collaboration on AI risks. “China has some of the best AI talent in the world,” he said. “I really hope Chinese AI researchers will make great contributions here.”
AI
-
LLama Adapter Development Benchmarked Against ChatGPT
By
–
yes, but we're working on an adapter on top of LLama too. Something we hope will be decently benched vs chatgpt.
-
Llama and QLoRA: The Cost-Effective Path to LLM Competition
By
–
Also don't forget llama exists and weights are available all over the internet. It's illogical to build an LLM from scratch if your goal is build a Chatgpt competitor. Qlora exists to reduce compute costs. Cost argument is weak unless there's some need to reinvent the universe.
-
CUDA Setup Challenges on Runpod GPU Infrastructure
By
–
somehow the corners of Twitter have made it about IITs and VC and something something. What you really need is to suffer with setting up the cuda version you *actually* want on Runpod.
-
True builders focus on shipping, not nationalism
By
–
People read too much into what Sam said and now topping it up with a little bit of nationalism. True builders build for fun and the joy of trying to ship something new, improving over shitty iterations. Pseudo builders make it about the nation. Just ship, folks. Tune out.
-
Sales Experience at Musigma and Fractal Analytics
By
–
anyone worked in sales at musigma/fractal analytics? want to chat.
-
AGI Safety and Governance Speaking Tour Across Asia
By
–
thanks to the @BAAIBeijing for inviting me to speak (virtually) about the importance of AGI safety and governance; really enjoyed! now off to japan, singapore, indonesia, and australia (then home ) trip has been amazing; incredible talent and energy focused on AI everywhere
-

Instruction Tuning Capabilities on Open Source Resources
By
–
How Far Can Camels Go? Exploring the State of Instruction Tuning on Open Resources Wang et al.: https://
arxiv.org/abs/2306.04751 #ArtificialIntelligence #DeepLearning #MachineLearning -

Leadership in Tech: 4 Pathways for Women Advancement
By
–
#Leadership in Tech: Here are 4 pathways to getting more women to the top https://
bit.ly/3XwJmsK via @wef #WomeninTech -
Red-teamers can likely make AI models hallucinate
By
–
I'd be very surprised if the red-teamers can't make it hallucinate. That's much stronger than "very few ordinary people run into it".