If, at the end of the year, we get a “Word/Phrase of the Last Half-Decade”, it’s pretty likely that it’s going to be “AI”.
OpenAI, the brand behind DALL-E, a text-to-image model, Sora, a text-to-video model, and ChatGPT, is taking the digital world by storm.
In fact, OpenAI is currently in talks with multiple companies in its latest funding round that would value the startup at more than $150 billion. Thrive Capital plans to invest $1 billion, with Microsoft, Apple, and Nvidia planning to join and provide funding, too.
The company was valued at around $80 billion at the start of this year and roughly $29 billion in 2023. With its valuation growing by $121 billion in just over a year, it’s clear OpenAI is leading the charge towards mass AI implementation.
The startup has been rapidly developing ever since it released its ChatGPT chatbot in November 2022. With its products available for both personal and professional use, you can’t really wonder why its weekly active users have grown to over 200 million, double the number it had last November.
For context, X reportedly has over 200 million daily users, so ChatGPT is going toe-to-toe with the big guns in terms of regular, active users. Granted, it’s not at the same level just yet, but give it time; it’s likely to get even bigger.
And so, that leads us to September 2024. OpenAI has just introduced its latest update to its ChatGPT model, one that’s set to make it even smarter than before.
ChatGPT and Its Brand-New Model
o1 is the name of the preview form given to “Strawberry,” OpenAI’s brand-new, super smart chat model that’s all about reasoning.
“Wait, reasoning? It’s an AI model. How does that work?”
Well, according to OpenAI, Strawberry is designed to spend more time thinking before it responds. So, it’s an AI model that will literally think before it gives you an answer. That sounds a lot like having a conversation with a human being, doesn’t it?
In recent tests, this new model performed similarly to PhD students on challenging benchmark tasks in physics, chemistry, and biology. Also, in a qualifying exam for the International Mathematics Olympiad (IMO), GPT-4o supposedly correctly solved only 13% of problems, while the brand-new reasoning model scored 83%.
Because of the new-found level of AI capability and the significant advance in AI tech that Strawberry represents, OpenAI is reversing the counter back to 1 and naming this series OpenAI o1.
Unsurprisingly, because this model is far more advanced than anything previously, OpenAI has had to devise a new safety training approach that aims to ensure that o1’s reasoning capabilities adhere to safety guidelines.
One way OpenAI measures safety is by how its models follow safety rules when someone tries to bypass them (known as “jailbreaking”). On its most complex tests, GPT-4o scored 22/100, while o1-preview scored 84/100.
OpenAI has also formalised agreements with the US and UK AI Safety Institutes to give them early access to a research version of o1-preview, enabling them to proof the models before public release.
OpenAI is also expected to charge a significantly higher price for the next generation of its AI models. OpenAI executives are supposedly considering charging $2,000 (this could be monthly, yearly, or a flat fee; it’s not confirmed yet) for access to the most advanced tools. In comparison, ChatGPT Premium costs $20 a month, which gives access to GPT-4o, its current flagship model.
We’re quite intrigued by how o1 is going to transform the AI market. o1 may not have any use for us, but who knows? Let’s see what happens over the next few months.
Feature Photo by Jonathan Kemper on Unsplash
Header Photo by Maxim Berg on Unsplash
.svg)