This feeds into one of the key trends I predicted for 2023: the evolution of open-source initiatives to democratise (Gen) AI.
INNOVATION
-
AWS and Hugging Face collaboration prevents AI monopolies
By
–
I’ve contemplated a future where a few well-resourced actors own and control powerful AI systems – and I have to say, this doesn't necessarily strike me as a good idea. Collabs like this between AWS and @huggingface are going help prevent AI monopolies.
-
Hugging Face AWS collaboration makes generative models widely accessible
By
–
Why the @huggingface / @awscloud collab is a bid deal: Hugging Face is on a mission to make the best generative models (which require a lot of data, computing power and basically $$$$$) widely accessible.
-
Meta Releases LLaMA Foundational Language Model Publicly
By
–
Today we're publicly releasing LLaMA, a state-of-the-art foundational LLM, as part of our ongoing commitment to open science, transparency and democratized access to new research. Learn more & request access https://
bit.ly/3SoXdQE -
LLaMA: Competitive AI Model That’s Smaller and More Efficient
By
–
LLaMA (Large Language Model Meta AI) achieves results competitive with the best currently released models while being smaller & more efficient — increasing accessibility to this technology for more researchers working on this important subfield of AI across the globe.
-
AI Platform Accelerates Business Value and Revenue Growth
By
–
We are committed to providing an AI Platform helping teams deliver faster time to value, increased revenue, and reduced costs. How do we do it? Register for our March 16 virtual event to find out more. #AIVisiontoValue https://
bit.ly/3KeMdU3 -
LLaMA-65B Outperforms Chinchilla and PaLM on Reasoning Benchmarks
By
–
On Common Sense Reasoning, Closed-book Question Answering, and Reading Comprehension, LLaMA-65B outperforms Chinchilla 70B and PaLM 540B on almost all benchmarks.
4/n -
Open Dataset Approach: LLaMA’s Reproducible Alternative to Chinchilla and GPT-3
By
–
Unlike Chinchilla, PaLM, or GPT-3, we only use datasets publicly available, making our work compatible with open-sourcing and reproducible, while most existing models rely on data which is either not publicly available or undocumented.
2/n -
Models trained on 1T tokens with continued improvement at 7B scale
By
–
All our models were trained on at least 1T tokens, much more than what is typically used at this scale.
Interestingly, even after 1T tokens the 7B model was still improving.
3/n