I think what was meant is that you actually don’t need huge models to have great output. I mean, Mixtral8x7b easily beats 70b to 100b models. It all depends on the way models are trained and the trainingdata. And for the future years we need SMLs for having them on devices
RESEARCH
-
AI-Generated Images: Copyright and Unauthorized Use Issues
By
–
I generated that with the GPT dalle feature. It’s totally unique. These people are just downloading my image and using it
-
Personal Philosophical Responsibility in Technology Ethics
By
–
philosophical responsibility is on oneself, not what others choose to do
-
OpenAI Announcements Lack of Pre-Briefing Access Criticism
By
–
I'd love to offer some more technical context where applicable on the OpenAI announcements today but, alas, I have not been pre-briefed like everyone else posting their stories at 9 A.M. PST. One day!
-
Kaiming He’s Deep Residual Learning Impact on AI
By
–
Deep Residual Learning and How It Shaped the AI Landscape – Kaiming He Finally got to watch the talk of Kaiming He, the lead of some of the most influential works in computer vision such as ResNet, Mask R-CNN, MAE(masked autoencoders), and contributor to other works that shaped
-
Speculative Decoding for Whisper: Notebook and Blog Guide
By
–
Notebook: https://
colab.research.google.com/github/sanchit
-gandhi/notebooks/blob/main/speculative_decoding.ipynb
… Blog Post: https://
huggingface.co/blog/whisper-s
peculative-decoding
… -
AI Product Evals Improvement and User Satisfaction
By
–
I appreciate the skepticism, but this is not true. We now have much more comprehensive evals to track this type of behavior which as it turns out were not robust enough to begin with. We are pushing hard and want people to love using our products!
-
NIST AI Funding Strengthens U.S. Competition Innovation
By
–
Last April we shared a memo outlining specific AI programs within @NIST that would benefit from greater funding.
When NIST has set ambitious technical agendas in the past (e.g., via FRVT), it has strengthened U.S. competition and innovation. -
Let’s Go Shopping: Web-Scale Image-Text Dataset for Visual Concepts
By
–
Let's Go Shopping (LGS) — Web-Scale Image-Text Dataset for Visual Concept Understanding Bai et al.: https://
arxiv.org/abs/2401.04575 #ArtificialIntelligence #DeepLearning #MachineLearning -
Small Models Will Have Huge Impacts, apples_jimmy Was Right
By
–
So @apples_jimmy was right from the beginning, that its the small models that will have huge impacts.