Llama 3 ANNOUNCED by Meta & Mark Zuckerberg AI NEWS

#largelanguagemodel #ainews #artificialintelligence #generativeai #meta

Llama 3" has just been announced by META, MARK ZUCKERBERG.

First of all, what is Llama? Llama 3 is an accessible, open-source large language model (LLM) just like ChatGPT by OpenAI, designed for developers, researchers, and businesses to build, experiment, and responsibly scale their generative AI ideas.
At numbers 2 and 3, I would like to share Llama 3's benchmarks and training data.
It was trained on 15 trillion tokens of data. A token represents basic individual words ("chatbot") or sentences ("How may I assist you today?"). So, Llama 3's data was seven times larger than that used for Llama 2, and it includes four times more code. This model has an 8k context length, which is double that of Llama 2, but ChatGPT has a 128k context window. So here are the benchmarks: Llama outperforms in 8 billion and 70 billion against Google's Gemma, Mistral, and Claude.

Organizations can get Llama 3 by requesting access on the official website.
It's an open-source model, so the code for all Llama models is available on GitHub.
Meta has also released a comprehensive set of open-source cybersecurity safety evaluations for large language models. It measures an LLM’s susceptibility to prompt injection, automated offensive cybersecurity capabilities, and propensity to abuse a code interpreter. They are also investing in safety and trust guides, from determining use cases to evaluating and improving safety to final development.
Code Llama has the potential to make workflows faster and more efficient for current developers and lower the barrier to entry for people who are learning to code.
Meta's AI assistant, which you can use in feeds, chats, and search across Meta's app:
Make Meta AI work for you.
Search integration.
AI in feed.
Imagine features.

#machinelearning
#deeplearning
#neuralnetworks
#robotics
#datascience
#googledeepmind
#grok
#openaisora
#samaltman
#claude3
#mistrall