This week marks a historical moment: The launch of the @NSF pilot of the National AI Research Resource! #NAIRR This will strengthen America’s position as a leader in AI and democratize AI resources for public sector AI innovation. https://
new.nsf.gov/news/democrati
zing-future-ai-rd-nsf-launch-national-ai
… 1/n
RESEARCH
-
NSF Launches National AI Research Resource Pilot Program
By
–
-
Stanford HAI leads national cloud computing initiative with universities and tech
By
–
.
@StanfordHAI has been championing this effort since we formed in 2019. We organized universities and tech companies to support this endeavor in 2020. https://
nytimes.com/2020/06/30/tec
hnology/national-cloud-computing-project.html
… 2/n -
U.S. Government Task Force for National AI Research Resource
By
–
We also called for the creation of a U.S. government-led task force to establish a NAIRR. “Support from Congress and the President could have a meaningful impact on American innovation,” said @StanfordHAI Co-Directors John Etchemendy and @drfeifei
. https://
hai.stanford.edu/news/national-
ai-research-resource-ensuring-continuation-american-innovation
… 3/n -
LLM Evaluation Methods and Gemini Pro Performance on LMSYS
By
–
Evaluation of LLMs is very hard and nuanced (especially academic evals which are leaked massively). Evals that rely on human judgement are far superior, so it feels good that Bard Gemini Pro (free tier) climbed pretty high on lmsys Looking forward to Gemini Ultra release!
-
Accelerating Matrix Multiplication for LLM Efficiency
By
–
Most of an LLM’s memory & compute are consumed by matrix multiplication operations. Today on the blog, learn about techniques used to accelerate mixed-input matrix multiplication for increased efficiency w/ performance close to peak hardware capabilities ↓
https://
goo.gle/3tXPuS8 -
AI Protein Design and DNA Synthesis: Navigating Biosecurity Challenges
By
–
With transformer #AI protein design and major advances in DNA synthesis, how do we deal with biosecurity concerns?
Editorial @ScienceMagazine https://
science.org/doi/10.1126/sc
ience.ado1671
…
David Baker @UWproteindesign and @geochurch -
Ludwig v0.9.3 Release: Phi-2 Support and LLM Improvements
By
–
Announcing #Ludwig v0.9.3! Support for microsoft/phi-2 Ensure correct padding token for #Phi and Pythia models Cast LLMEncoder output to torch.float32, freeze final layer at init Enable IA3 adapters Add batch size tuning for #LLMs https://
pbase.ai/3HAOxCs -
SnorkelAI Launches 7B Model Achieving SOTA AlpacaEval 2.0
By
–
Get your hands on the new 7B model that put @SnorkelAI SOTA on AlpacaEval 2.0! This work is foundational to programmatic alignment support in Snorkel Flow for steerable LLMs w/out manual preference annotations. Download, sandbox, or API calls http://
snkl.ai/sm -
Testing Economy Model for AI Safety and Innovation
By
–
"The best way to solve these [#AI] problems is to create a new set of testing companies that will be incentivized to out-innovate each other—in short, a robust economy of testing."—
@ericschmidt -
Google Gemini Pro Outperforms GPT-4 OpenAI Responds
By
–
Google made massive improvements to Bard, with Gemini Pro now ranking higher than regular GPT-4. Keep in mind, Gemini Pro is Google's SECOND best model. We're still waiting on Gemini Ultra. Expect OpenAI to drop a new model anytime now to stay in the lead.