#A.I. Import AI publishes first on Substack – subscribe here. Cambridge researchers show how to use distributed training to make a 1.3bn parameter LLM:…More evidence that distributed training works well for relatively small models…Researchers with ...

#A.I. Import AI publishes first on Substack – subscribe here. FlashAttention-3 makes it more efficient to train AI systems:…Significant efficiency improvements…Researchers with Colfax Research, Meta, NVIDIA, Georgia Tech, Princeton University, and ...

#A.I. Import AI publishes first on Substack – subscribe here. Google beats a hard security benchmark, showing how unoptimized today’s LLMs are:…Take one LLM plus some well-designed scaffolding and a hard benchmark gets ‘solved’…Google has published...

#A.I. Import AI publishes first on Substack – subscribe here. Microsoft shows that human-level voice cloning is here:…VALL-E 2 portends the wild synthetic voice future – but it’s not being released (yet)…Microsoft has made further progress in text-...

#A.I. Import AI publishes first on Substack – subscribe here. Why are people skeptical of AI safety policy?…A nice interview with the Alliance for the Future…Here’s a good interview with Brian Chau, a founder of the DC-based AI advocacy group Allia...

#A.I. Import AI publishes first on Substack – subscribe here. Chinese researchers build a hard benchmark for multimodal understanding:…Visual LLMs still struggle with localization and complex visual reasoning…Chinese researchers have introduced MMT...