Describing 2023 as a transformative year for Artificial Intelligence (AI) would be an understatement. AI transitioned from a mere buzzword symbolizing the distant future to becoming seamlessly integrated into our daily lives. OpenAI’s ChatGPT emerged as a central figure in the AI discourse. During OpenAI’s DevDay event last November, CEO Sam Altman revealed that ChatGPT boasted an impressive 100 million weekly active users, accompanied by a robust developer community of two million individuals leveraging its API. OpenAI remained agile throughout 2023 by introducing GPT-4, adding an App Store featuring third-party plugins, releasing the capability to craft customized GPTs, and various other advancements.
However, the journey was not without its challenges. In mid-November, Sam Altman faced a brief termination, followed by his reinstatement after ten days. Adding to their complexities, news emerged just before the new year that OpenAI was in a legal dispute with the New York Times over alleged unauthorized use of copyrighted content. The outcome of this legal matter holds significant implications for OpenAI and the broader landscape of Large Language Models (LLM) for the foreseeable future.
If you are someone that has heard of an LLM before but possess only a limited understanding of what it means, let’s dig a little deeper. An LLM is a program that understands and uses language in a way that is similar to how humans communicate. It’s trained by ingesting a massive amount of text from the internet, books, and other sources, learning the patterns and meanings of words and sentences. An LLM effectively recognizes patterns in language which allows it to generate coherent and contextually relevant text in response to specific requests. Its capacity for knowledge can continue to grow by adding new data. Furthermore, the LLM can undergo pretraining in specific domains to yield even better results.
While initially focused on text, LLMs have expanded their capabilities over time to include inputs such as images, audio, and video. This enables the LLM to process and generate content in a more diverse and comprehensive manner. These new models are called multimodal large language models. We are in the early stages of these tools, but advancements are happening daily, and the rapid pace of change is remarkable.
In a world where companies invest billions in developing LLMs, concerns arise regarding the black box technology they employ. LLM search queries come at a considerable cost, demanding upwards of ten times more processing power than standard searches. At scale, the operational expenses for running these searches can soar into the millions. With certain proprietary LLMs offering complimentary usage, the old saying goes, “If you are not paying for the product, you are the product.” This has prompted some to explore alternative approaches.
This mindset has spurred a movement towards leveraging open-source LLMs. The numerous advantages in adopting an open-source LLM include heightened privacy, customizable models, transparent code, and community support. Surprisingly, you can run these open source LLMs locally on your own device or on a cloud instance. Depending on the LLM and the size of the dataset, they require different resources. A quick search online will instruct you on how to run an LLM locally. If you have spare time and wanted to experiment, here are a few open-source options in no particular order:
- LLaMA 2 – Large Language Model Meta AI
- BLOOM – BigScience Large Open-science Open-access Multilingual Language Model
- Falcon – Technology Innovation Institute
- Mistral – Mistral AI
Setting up and running your own LLM locally can be a fun and engaging exercise. However, once operational, it’s crucial to acknowledge that no LLM is entirely free of bias and misinformation. Therefore, it is imperative to verify any results against trusted sources. While engaging in this activity might be interesting in the short term, it’s highly probable that, in the not-too-distant future, LLMs will run locally on your phone or tablet.
The future has arrived, and the pace of change shows no signs of slowing down. However, the looming New York Times vs OpenAI trial mentioned earlier poses a major disruption for LLMs. Considering their extensive training on copyrighted material, a verdict against OpenAI has the potential to create a significant ripple effect, impacting other LLMs as well. With the genie now out of the bottle, it will be challenging to return it, but it will certainly be intriguing to see how this situation plays out.
Regardless of the trial outcome, your essential work persists. Genuine research conducted by scholars and students will continue to rely on the steadfast support of librarians and libraries every step of the way.
Enjoying the Atla Blog?
Subscribe to receive email alerts of new blog posts of a specific type. Members, subscribers, publishers, or anyone interested in the study of religion & theology are welcome to sign up to one or all alerts to keep up to date with the Atla community. If you or your institution are a member, the Atla Newsletter delivers a monthly curated email of top posts to your email inbox.