Meta’s LLaMA Model Has Been Leaked – Here’s What It Means For You

If you’ve been following the world of AI text generation these last few months, there has been a lot of exciting new developments!

One of those is the release of Facebook’s new line of models, LLaMA. While these were originally intended to only be available to researchers, they were leaked as of March 2nd, 2023, and are now available for anyone to download.
On this page, I’ll explain why this is so significant, what makes LLaMA different, and how you can use it to make your smut generating even better!

What Is LLaMA?

LLaMA stands for Large Language Model Meta AI, and it was a set of language models created by Meta, the company that owns Facebook.

If you’ve heard of ChatGPT in the news recently, it works kind of the same – you can give the model a prompt, and it will generate new text back for you.

You can currently see large language models in use on OpenAI’s website, or even start generating smut (and engaging in other naughty fantasies) through services like NovelAI. ChatGPT does NOT allow for adult content (or anything remotely unsavory), so at the moment, smaller models are all that are available.

Because large language models use such a large amount of computing power, they are very difficult to run on consumer hardware – even if you have a high-end PC.

Thankfully, LLaMA may just bring the hope of generating content on our own computers – without the use of a third-party service, closer than ever.

What Makes LLaMA Special For Generating Adult Content?

Currently, there are two things that make the leak of LLaMA so significant.

The first, and perhaps the most significant, is that it performs extremely well for how small the models are.

One way models are measured is in number of ‘parameters.’ You can sort of think of this as how ‘smart’ the AI is – the more parameters, the more capable it is of appearing intelligent and generating quality responses. However, as the number of parameters increases, the hardware requirements to use it go up dramatically.

The problem is, depending on how effectively the model is trained, it can have a drastic effect on performance.

LLaMA’s performance benchmarks show that it may perform better than models 10x its size on many tasks even matching OpenAI’s GPT-3. In a nutshell, this means that you can run a way more powerful model, with less powerful hardware.

Considering that the 13 billion parameter version of LLaMA can run on today’s high-end consumer graphic cards, this is significant news!

Additionally, the second point LLaMA has going for it is that it appears to be unfiltered – it will generate content on any topic, without trying to block, censor, or redirect outputs.

This point isn’t quite as significant since people will fine-tune censored topics back into it anyway, but it gives people a good starting point out of the gate, and may lead to higher quality outputs after it’s finetuned than if it started out censored.

What We Can Expect Going Forward

LLaMA is already available for public use, through The Horde on KoboldAI. Simply click on the ‘AI’ button, and select any of the ‘LLaMA’ options that may be available.

However, these models don’t typically reach their full potential out of the gate.

It will take some time to figure out the best settings to use for each model, and there may be adjustments to how they work that lead to better outputs going forward.

Additionally, it’s possible that somebody will use LLaMA as a base for new finetunes. For example, the ‘Erebus’ model has been finetuned specifically on erotica, so it punches above its weight for this sort of content.

Closing Thoughts

It’s a very exciting time for AI text generation!

Not only can we expect computer hardware to continue getting more powerful, but we’re getting better and better models, even with the same hardware requirements.

I’m hoping within a few years, even a non-enthusiast will have hardware powerful enough to run language models locally.

Will that happen? Time will tell!

In any case, be sure to return back here to stay up-to-date with all news and updates.

– Lexi