MACHINE LEARNING - AI Dynamics

Open Source AI History Before Llama and PyTorch

By

@ylecun

–

16 December 2023 14h55

Open source AI didn't start with Llama, nor with PyTorch.

→ View original post on X — @ylecun

16 December 2023

Reinforcement Learning for Diffusion Models from Scratch

By

@miketamir

–

16 December 2023 3h01

Dr. Tanishq Abraham – Reinforcement Learning for Diffusion Models from Scratch https://
bit.ly/3GuPUBM #AI #MachineLearning #DeepLearning #LLMs #DataScience

→ View original post on X — @miketamir

16 December 2023

Google Gemini Pro Shows Significant Improvements in Understanding and Reasoning

By

@poe_platform

–

16 December 2023 0h54

Internal performance evaluations from Google have shown significant improvements for Gemini Pro over previous Google models in understanding, summarization, reasoning, coding, and planning. This model also excels in cross-modal reasoning. (2/3)

→ View original post on X — @poe_platform

16 December 2023

Latent Variables: Parameters vs State in Gradient Descent

By

@ylecun

–

16 December 2023 0h50

Okay so the latent variable you infer through gradient descent look like parameters, not like state variables.
But why?

→ View original post on X — @ylecun

16 December 2023

ML Advances: TpuGraphs Dataset and Graph Segment Training

By

@googleai

–

15 December 2023 23h53

Read all about recent advances in using ML to improve efficiency of ML workloads — including TpuGraphs, a performance prediction dataset on large tensor computational graphs, and Graph Segment Training, a novel method to scale graph neural networks →
https://
goo.gle/3RrnQUZ

→ View original post on X — @googleai

15 December 2023

Mixtral: The Future of Efficient Language Models

By

@mreflow

–

15 December 2023 22h53

I think the way Mixtral works is likely the future of language models. It seems more efficient and scalable in my opinion.

Here's a quick oversimplification of how it works that I clipped from today's news video (on YouTube): pic.twitter.com/zIZXjxY4fk
— Matt Wolfe (@mreflow) 15 décembre 2023

I think the way Mixtral works is likely the future of language models. It seems more efficient and scalable in my opinion. Here's a quick oversimplification of how it works that I clipped from today's news video (on YouTube):

→ View original post on X — @mreflow

15 December 2023