Meta Llama 3 Review + Use Cases

By Maryam FarooqTop Resource

Highlights

🤩

Original Resource Link: Meta AI Blog & Andrej Karpathy’s post on X
Original Resource Publish Date: April 18, 2024
Original Resource Author: Meta AI & Andrej Karpathy

  • The Meta AI team released Llama 3, “the most capable openly available LLM to date”, on April 18th 
  • The 8B & 70B models come in pre-trained and instruction tuned variants 
  • Andrej Karpathy, previously Director of AI at Tesla & on the founding team of OpenAI, shares his initial thoughts around tokenizers, architecture, sequence length, training data, scaling laws, and systems 

Notes on the Resource

📝

The Meta AI team released Llama 3 on April 18th – according to them, “the most capable openly available LLM to date”. The 8B & 70B models include new capabilities such as improved reasoning, and come in pre-trained and instruction tuned variants.

They developed a new high-quality human evaluation set that contains 1,800 prompts that cover 12 key use cases including:

  • Asking for advice
  • Brainstorming
  • Classification
  • Closed question answering
  • Coding
  • Creative writing
  • Extraction
  • Inhabiting a character/SL Persona
  • Open question answering
  • Reasoning
  • Rewriting
  • Summarization

Llama 3 also encodes language much more efficiently by using a tokenizer with a vocabulary of 128,000 tokens which leads to substantially improved model performance.

Impressions 🤔

Andrej Karpathy, previously Director of AI at Tesla & on the founding team of OpenAI, share a great overview around Meta’s Llama 3 release on April 18th covering his initial thoughts around tokenizers, architecture, sequence length, training data, scaling laws, and systems 

He points out that there is also some more interesting information on the model card here.

In the less than 2 months Llama 2 has been out, we’ve seen an interesting range of use cases built. Including the 3 winners from AI at Meta’s #Llama3Hackathon that took place on May 14, 2024:

  1. Open Glass AI: Open source smart glasses built with $20 of parts
  2. Deb8: An AI agent debate arena
  3. Team AAA: Discovery of new jailbreaking methods for Llama 3

What's Missing?

🕵️

Have you played around with Llama 3? What do you think? Email me, and let me know your thoughts.

 

Maryam
maryam@nyai.co

maryam-farooq

Maryam Farooq

Founder, NYAI

Subscribe to Substack

Maryam Farooq is a community builder + frequent speaker on AI, and an early-stage startup advisor. She currently serves as the Founder & Director for NYAI (New York Artificial Intelligence). She is former Co-founder & COO of generative AI startup Aggregate Intellect, one of 7 companies to graduate the CDL Toronto AI program in 2023.