2024 was a groundbreaking year for AI and Large Language Models ("LLMs"), reshaping how we think about technology's role in our businesses (and lives). The IoD Next Gen Committee looks back on 2024 to share what it thinks are some of the biggest events in this space and why.
It feels right to start with OpenAI’s year, considering that it was their ChatGPT tool which first truly captured our attention... along with millions of other people! (Do you remember when ChatGPT got to 1m users in only 5 days in 2022?!)
These are the models and tools we have seen specifically from OpenAI that we think are incredibly useful and versatile to today's businesses and beyond:
When? | What? | Why noteworthy? |
May 13, 2024 | GPT-4o | A multimodal model, this allows users to process and generate text, images and audio, enabling more natural and versatile interaction with the model.
Also, this model supports a huge context window, facilitating complex interactions. It is also twice as fast and half as expensive to run as its immediate predecessor, GPT-4 Turbo.
This was quite a big step up for users in terms of speed, quality of responses, and working with images and files. |
Sep 28, 2024 | Advanced Voice Mode | This feature enables users to engage in real-time, human-like conversations with the AI, enhancing the interactivity and accessibility of ChatGPT.
This tool is great for “discussing” ideas or concepts and actively surrounding a problem from multiple perspectives to look for what you don’t know. |
Oct 03, 2024 | Canvas | This was a new interface for working with ChatGPT on writing and coding projects that go beyond simple chat. Canvas opens in a separate window, allowing you and ChatGPT to collaborate on a project.
This tool is fantastic for working on blocks of code or papers or emails. Anything where you’re writing up a good chunk of text in iterations. |
Dec 05, 2024 | o1 | Designed to handle complex, multi-step tasks with advanced accuracy, o1 employs a "chain-of-thought" reasoning approach, breaking down problems into smaller logical steps. o1’s benchmarking scores really caught our attention as the model significantly outperforms previous models (and, in some cases, PHD level humans) in areas like advanced mathematics, coding, and scientific reasoning. This is a big deal. |
OpenAI has been up to a lot more as well with their text-to-video tool Sora (not yet available in Guernsey) and Chat GPT search.
The interesting thing I have found in following OpenAI - at least in their capacity as a core frontier player in this space - is that they continue to be very good at showing us all the different ways that we will be able to interact with LLM-powered tools. They started with ChatGPT and continue to develop and show us different ways of interacting with these models whilst still improving the calibre of the underlying models themselves.
In the open source space, we have seen lots of exciting developments as well. I won’t even attempt to summarise what’s been going on in the enormous world that is “open source”, but certainly there are some fantastic models coming out from organisations like Meta and Mistral. Meta's Llama 3.2, released in September 2024, brought significant improvements in speed and accuracy, making it a favourite among researchers for both academic and practical applications. It is particularly interesting that Meta has decided to commit to releasing their Llama suite of models as open source. Mistral released Mistral Large 2 in July 2024 which introduced a new level of efficiency and versatility, offering exceptional performance in tasks ranging from natural language processing to code generation. Its architecture prioritizes modularity, making it easier for developers to customize and integrate into diverse workflows, further solidifying its place as a standout open-source model this year. There are some fantastic models out there that we can, today, implement in our own businesses and apply to countless use cases.
Anthropic’s public beta release (in October 2024) of a new “computer-use” feature as part of their upgraded Claude 3.5 Sonnet model definitely attracted our attention. This functionality enables Claude to interact with a computer’s graphical user interface similarly to a human user - interpreting screen contents, moving the cursor, clicking buttons and typing text. It feels to us like this might be a “big deal” too.
I'm sure you will agree that a lot has happened in 2024! Going into 2025, the IoD Next Gen Forum is committed to delivering workshop opportunities, to unpack the tools above in a bit more detail, and understanding what these tools mean for us, as future business leaders, and in our businesses in today's landscape. We are asking questions like: What does the team of the future look like? What tools should we be adopting now? What risks does this give rise to?
We would like to bring the Next Gen Forum on this journey with us in truly understanding LLM-powered tools and what they mean for our businesses and future boards. Make sure you hear about our upcoming workshops and events through our mailing list, which you can sign up to here.
Our first workshop of the year will be in this space, and we are exciting to release tickets and further details in the coming weeks.