Turkish subtitles - 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.
Subtitles generated on my computer; they follow the video — click a line to jump there. You must make your decision by evaluating the differences between large-scale adopted models and those with accessibility and customizability. This video is published on youtube.com; the player here is YouTube’s. The Turkish summary and subtitles are prepared for AiPulse. Watch on YouTube AI Critique → |