Published On Jul 24, 2024
Llama 3.1 is here, and if anything, it’s paper is even more impressive. It’s like Meta want to reveal the secret sauce of LLMs. I go through the highlights of all 92 pages and test Llama 405B on the SIMPLE bench, my new private, vetted general intelligence benchmark, against GPT 4o and Turbo, Gemini 1.5 Pro and Claude 3.5 Sonnet.
Weights and Biases Link: https://wandb.me/ai_explained
AI Insiders: / aiexplained
Llama 3.1 Paper: https://scontent-lhr6-2.xx.fbcdn.net/...
Llama 405B Web Release: https://ai.meta.com/blog/meta-llama-3-1/
Zuckerberg Money Run Out Interview: • Mark Zuckerberg on Llama 3.1, Open So...
Zuckerberg Llama 4: • Meet The New Mark Zuckerberg | The Ci...
OpenAI Losses: https://www.theinformation.com/articl...
Nvidia Nims: https://build.nvidia.com/explore/disc...
Open Source Definition: https://opensource.org/deepdive/draft...
Data Disappearing: https://www.nytimes.com/2024/07/19/te...
Infinite Bench: https://arxiv.org/pdf/2402.13718
ARC-A12 Challenge: https://www.semanticscholar.org/reade...
Adversarial Examples: https://arxiv.org/pdf/1707.07328
‘Systemic Risk’: https://x.com/deanwball/status/181582...
Zuckerberg-Hawley Letter: https://www.hawley.senate.gov/hawley-...
Groq Demo: https://x.com/JonathanRoss321/status/...
Altman Letter: https://www.washingtonpost.com/docume...
Lmsys Response: https://x.com/lmsysorg/status/1816010...
Scale Private Leaderboard: https://scale.com/leaderboard
AI Insiders: / aiexplained
Non-hype Newsletter: https://signaltonoise.beehiiv.com/
GenAI Hourly Consulting: https://www.theinsiders.ai/