There are over 325,000 models on Huggingface and
thousands more are being added. And why might you choose to use AI models like these? Well,
let's start by getting a few things straight. The models we're talking about in this video,
they're specifically LLMs, and that's large language models, which are foundation models
that use artificial intelligence, deep learning and massive datasets to generate text. We're
talking generative AI. And there are two types of generative AI models: There's proprietary models,
and there are open source models. Now, proprietary LLMs, those are owned by a company who can control
its usage. A proprietary LLM may include a license that restricts how the LLM can be used. On
the other hand, open source LLMs are free and available for anyone to access, and developers and
researchers are free to use, improve or otherwise modify the model. Now look, it's not true in every
instance, but generally many proprietary LLMs are far larger in size than open source models. And
specifically in terms of parameter size. Some of the leading proprietary LLMs extend to thousands
of billions of parameters. Probably? Actually, we don't necessarily know because, well, those
LLMs and that parameter counts are proprietary. But bigger isn't necessarily better. And the
open source model ecosystem is showing promise in challenging the proprietary LLM business model.
So let's discuss the benefits of open source LLMs. Let's talk about the types of organizations
that are using them. Let's talk about some of the leading open source models available today,
and we should talk about the risks associated with using them. Now, clearly, one of the
benefits of a open source large language model, that has to be transparency. Open source LLMs
may offer better insight into how they work, their architecture, and the training data used to
develop them. Another big one is pre-trained open source LLMs allow a process called fine tuning.
That means you can add features to the LLM that benefit your specific use case and the LLMs
can be trained on specific data sets. So I can fine tune an LLM with my own data. And community
contributions are a big plus. Using a proprietary LLM means you're reliant on a single provider,
whereas open source models benefit from community contributions and multiple service providers. You
can experiment and use contributions from people with varying perspectives. And these benefits
have led to all sorts of organizations to use open source LLMs. In another video, I addressed
how NASA and IBM developed an open source LLM trained on geospatial data. Some healthcare
organizations use open source LLMs for diagnostic tools and treatment optimization. There's an open
source LLM called FinGPT [fin / financial]. It was developed for the financial industry. Which brings
us onto the topic of talking about some specific open source LLMs that you might find of interest.
Now Huggingface maintains an open LLM leaderboard, and that tracks , ranks, and evaluates open
source LLMs on various benchmarks like which LLM is scoring highest on the Truthful AI Benchmark
series, which measures whether a language model is truthful in generating answers to questions. So
it gives those answers a score. And the top spots on this leaderboard, they change frequently.
And it's quite fun to watch the progress these models are making. Many of the models on the
leaderboard are variations on the Llama 2 open source LLM. That's the one provided by Meta
AI. And Llama 2 encompasses pre-trained and fine tuned generative text models from 70 billion
all the way down to 7 billion parameters. And it's licensed for commercial use. Vicuna was created
on top of the Llama model and fine tuned to follow instructions. And then it's also Bloom by BigScience, which is a multilingual language model created by more than 1000 AI researchers. Now, one
area that both proprietary and open source LLMs share is their associated risks. Although LLM
outputs often sounds fluent and authoritative, they can be confidently wrong. Hallucinations,
they can result from the LLM being trained on incomplete, contradictory, or inaccurate data
from misunderstanding context. Bias happens when the source of data is not diverse or not
representative. And security problems can include leaking PII, and cybercriminals using the LLMs for
malicious tasks like phishing. Especially in these early days of large language models, we do need to
mitigate risk. But open source LLMs are thriving in business. Here at IBM, the Watsonx.ai Studio
makes available access to multiple Llama 2 models, and IBM has released a series of foundation
models of its own called Granite. And this space is changing rapidly, making open source
LLMs a field well-worth keeping a close eye on.
