LLMS

→ View original post on X — @ninadschick

Elon Musk’s Grok3 Achievement Signals Bigger AI Role

By

@ninadschick

–

19 February 2025 21h50

What @elonmusk achieved with his team to build #Grok3 is a superhuman feat of engineering.

Could anyone else have pulled it off? Jensen Huang said it was ‘singular’.

That all means that Elon Musk will be a playing an even bigger role in frontier development of AI.

He… pic.twitter.com/tag6eY6De1
— Nina Schick (@NinaDSchick) 19 février 2025

What @elonmusk achieved with his team to build #Grok3 is a superhuman feat of engineering. Could anyone else have pulled it off? Jensen Huang said it was ‘singular’. That all means that Elon Musk will be a playing an even bigger role in frontier development of AI. He

→ View original post on X — @langchain

LangChain CEO Event Atlanta February 27th

By

@langchain

–

19 February 2025 20h54

LangChain in Atlanta! Join us next Thursday, February 27th for an evening of AI at the Honeywell offices with LangChain CEO, @hwchase17

→ View original post on X — @thom_wolf

Ultra-Scale Playbook: 5D Parallelism and CUDA Optimization Guide

By

@thom_wolf

–

19 February 2025 19h00

After 6+ months in the making and burning over a year of GPU compute time, we're super excited to finally release the "Ultra-Scale Playbook" Check it out here: http://
hf.co/spaces/nanotro
n/ultrascale-playbook
… A free, open-source, book to learn everything about 5D parallelism, ZeRO, fast CUDA kernels,

Phi-4 Model Release: 40% Synthetic Data Composition

By

–

19 February 2025 18h47

Phi-4! (
https://
arxiv.org/abs/2412.08905). 40% synthetic.

→ View original post on X — @sambanovaai

DeepSeek-R1 Game-Changer for Enterprise AI Deployment

By

@sambanovaai

–

19 February 2025 18h47

In his interview with @CliffSaran at @ComputerWeekly
, @RodrigoLiang explained why DeepSeek-R1 is a game-changer for enterprises aiming to deploy cutting-edge AI without the budget strain. Read the full story here: https://
computerweekly.com/news/366619398
/DeepSeek-R1-Budgeting-challenges-for-on-premise-deployments
… #AI

SFT and RL Training Pipeline: Limitations of Cold Start Approach

By

–

19 February 2025 18h30

Yes & no. They had SFT (cold start) → RL → SFT (CoT + knowledge) → RL. Not sure if they focused on providing solutions to hard problems via SFT CoT though, because that SFT data was just generated from the previous model (so if that model doesn't solve it, then you are stuck)

Alternating SFT and RL Training for Improved Model Policy

By

–

19 February 2025 18h24

I wonder if alternating SFT (with correct solutions) and RL can help here. I.e, SFT → RL → SFT → RL. SFT would help with improving the initial policy and help making the exploration less random perhaps, plus it expands/refines the search space?

Model Performance Comparisons: Size Normalization and Inference Scaling Challenges

By

–

19 February 2025 17h56

Model performance comparisons are tricky without knowing exact sizes; it could be an apples-to-oranges case. Plus, you can always push numbers with inference scaling. Normalizing by tokens/sec would be more meaningful.