Artificial Intelligence

Grok3 Beta Ranks #1 in CaseLaw Benchmark!

A lot has happened yesterday. Here is your full recap. Enjoy!

April 8th and 9th have been incredible in terms of AI News, Enjoy!

Big congratulations to Gashon Hussein from Stanford University, in collaboration with NVIDIA, on the release of your groundbreaking paper: "One-Minute Video Generation with Test-Time Training (TTT)" This is an exciting leap forward in generative video modeling. Generating one-minute-long Tom & Jerry cartoons, in a single pass with no editing or stitching, is no small feat. The temporal and spatial coherence you’ve achieved is truly impressive. We’re especially excited by the innovative use of TTT-layers on top of pre-trained Transformers. It’s clear this approach is opening up new possibilities in long-form video generation. Looking forward to seeing where this research leads next.

Exploring Generative AI's Impact on Business: A McKinsey & Company Report In their March 2025 report, "The State of AI: How Organizations Are Rewiring to Capture Value," McKinsey & Company delve into how large organizations are strategically leveraging generative AI to enhance their business models and operational efficiencies. Authored by Alex Singla, Alexander Sukharevsky, Lareina Yee, Michael Chui, and Bryce Hall, the study provides a comprehensive look at the transformative shifts occurring across industries due to AI integration. Key findings highlight the importance of restructuring workflows and strengthening AI governance to drive significant bottom-line impacts. The insights offered are essential for leaders looking to navigate the complexities of AI deployment and harness its potential for substantial economic gains.

A lot has happened from April 4th to April 7th, Make sure to stay in the loop. Enjoy!

Huge congratulations to the team at Meta on the release of Llama 4 – an incredible leap forward in open-source AI! We’re seriously impressed by what you've achieved here: A full re-design of the Llama series, introducing cutting-edge mixture-of-experts architecture and native multimodality The blazing-fast Llama 4 Scout (17B activated params, 10M+ token context, single GPU capable!) The high-performance Llama 4 Maverick, outperforming giants like GPT-4o and Gemini Flash – all while staying efficient and nimble. And the Llama 4 Behemoth preview – pushing boundaries on STEM benchmarks and setting a new bar for what’s possible in open-source frontier models Massive kudos to your team and your launch partners — Hugging Face , togethercompute, SnowflakeDB, Ollama, Databricks, and many more — for making this a reality. Llama 4 is a major milestone for open source AI, and we're excited to see how the community runs with it. The future is open, and it's looking very bright.

This was made for just $9 and in under an hour using Higgsfield’s new model.... keep up the great work Higgsfield AI!

Yet another newsletter is out! Enjoy!

Yesterday's AI News Recap is out! Enjoy!