What’s going on with LLMs? Beyond the idea that they are basically taking over so many processes, in business and beyond?

Well, we can look at some of the most powerful examples of these models, to know more about what they will look like really soon.

This year, our IIA event coincided with a really important roll-out of Meta’s Llama 3 LLM model. In one of our panels, you can hear Dave Blundin talking about this, and crediting Yann LeCun (who is in the audience) with the development of Optical Character Recognition, and looking at that in context, given what we have now.

I think it’s worth looking at what people can do with the new Llama 3 and what it brings to the table:

As Blundin pointed out, the system is trained with an astounding 15 trillion tokens, which is seven times larger than the last iteration.

It also adds something called group to query attention, which towardsAI characterizes as the “interpolation of multi-query and multi-head attention.” If that’s Greek to you, GQA has something to do with new ways to handle key-values. Techniques like weight only quantization and the use of prompts in PyTorch also helps with this type of use.

In terms of practical applications, Meta is unveiling its new Meta AI assistant that runs on the llama 3 model. As the company points out, the tech can help you to prepare meals, study for a test, etc.

It can also provide input on media feeds and bring in creativity with its ‘imagine’ feature.

That’s an example of what we do with these larger and more muscular LLMs.

What else?

Specifically, Meta identifies these 12 use cases –

· Asking for advice

· Brainstorming

· Classification

· Closed question answering

· Coding

· Creative writing

· Extraction

· Inhabiting a character persona

· Open question answering,

· Reasoning

· Rewriting

· Summarization

Quite a list!

And all of this is brand-new – we’ll see it getting a lot of use in the coming months. As Meta concludes, the system can “boost your intelligence and lighten your load, helping you learn, get things done, create content, and connect, to make the most out of every moment!”

Those with an ear to the ground largely predict llama 3 will soon “be everywhere.”

I think it’s a major example of practical steps forward that we’re taking…and worth thinking about right now.

Share.
Exit mobile version