This is huge. A new technique called Reflection-Tuning allows open-source models (Llama 3.1 70B) to outperform Claude 3.5 and GPT-4o. This new technique trains the model on structured, synthetic data to detect reasoning errors and enable LLMs to fix their own mistakes.

The Llama 3 paper is a must-read for anyone in AI and CS. It’s an absolutely accurate and authoritative take on what it takes to build a leading LLM, the tech behind ChatGPT, Gemini, Copilot, and others. The AI part might seem small in comparison to the gargantuan work on *data*

👇Crazy growth in the last 5 quarters for NVIDIA Data Center #AI #HPC via Beth Kindig

👇Quote tweeting this thread by V hoping someone from the larger #HPC community could help

Jack Dongarra at PPAM'24 in Ostrava. Still having fun in #HPC after 50 years with high-performance linpack, low performance HPCG, and exaflop/s mixed precision calculations. Onwards to low precision zettascale in the brave new #AI world! Looking forward to my keynote at 1:30pm.

US starts export controls for #quantum computers, related stuff, as well as Gate All-Around Field-Effect Transistor (GAAFET) Technology for #HPC chips that can be used in supercomputers. This is proposed for 2nm process technologies and below. eenewseurope.com/en/us-starts-e…

Today let’s remember Dennis M. Ritchie, who was born #OTD in 1941. Ritchie received the 1983 #ACMTuringAward with Ken Thompson, for their development of generic operating systems theory and specifically for the implementation of the UNIX operating system. bit.ly/3yqg0p8

Hatem Ltaief I opt for "The most efficient byte is the byte you don't transfer." Much more relevant performance and energy-wise IMO

#GenerativeAI will be a game changer for weather observation: Stan Forge, head of @Nvidia's #HPC program in Earth System Model & Computer Fluid Dynamics "It's expected that almost all institutions will use weather #AI models within the next 3 years" mk.co.kr/en/it/11113302

#AWS #HPC Blog: Science, not servers Customers wanted a cloud-native #HPC solution that combines the flexibility and scalability of Amazon Web Services with the familiar tools and workflows of traditional #HPC environments aws.amazon.com/blogs/hpc/you-…

Today would have been the 83rd birthday of Dennis Ritchie, inventor of C and co-creator of Unix: zd.net/2creeZi (Image v/SanderFocus, article v/Jason Perlow is on Mastodon, Bluesky, Threads & ZDNET)

.Oracle Cloud has “acres of @Nvidia #GPU Clusters for training large scale #AI models” #HPC #GenAI via Beth Kindig

HPC Guru Oracle Cloud NVIDIA Beth Kindig An 800MW DC to be filled with NVDA GPUs … that’s a few $$ … maybe this AI thing still has plenty of growth ahead :-)