Meta’s new version of large language model (LLM) Llama will be available no later than May.
The company announced this on Meta AI Day in London. “Within the next month, actually less, hopefully in a very short period of time, we hope to start rolling out our new suite of next-generation foundation models, Llama 3,” said President of Global Affairs Nick Clegg.
It is envisioned that Llama 3 will power multiple Meta products. The social media giant eventually wants to build a Llama-powered Meta AI that serves as the world’s most useful assistant. The LLM model launching in a few weeks is expected to have billions more parameters than its predecessor. The Llama 2 model has 70 billion parameters.
New capabilities
With these additional parameters, the LLM should be able to provide better answers. Llama 2 was Meta’s first foundation model to be publicly available, but it still received some criticism that it was too limited in its capabilities. Llama 3 is, therefore, expected to be able to respond correctly to many more prompts and queries. An earlier leak already gave an example of understanding the English question “how to kill a vehicle’s engine,” meaning how to shut it off rather than end its life.
Currently, Llama 2 has yet to leave competing models behind in benchmarks. In most tests, for example, GPT-3.5 and Grok-1 now score much better in language understanding and math. After the release of Llama 3, it will become clear whether Meta can step up the competition a bit.
Tip: Databricks releases DBRX: open-source LLM that beats GPT-3.5 and Llama 2