Thanks to the tremendous growth in popularity, AI and AI-powered tools are seeing a wave of investments and interests normally not seen in data science. But, according to OpenAI CEO, Sam Altman, the age of giant AI models has already come to an end. In a report by Yahoo, the CEO of OpenAI made these comments during a Zoom call with MIT. Saying in part, “I think we’re at the end of the era where it’s going to be these, like, giant, giant models,”.
With the rapid advance of competitors, such as Google’s Bard, or Databrick’s Dolly 2.0, it’s an interesting comment to make. But why? With this rapid shift in AI, why would Altman come to that conclusion? Well it’s simple, “We’ll make them better in other ways.”.
All of this comes as a number of well-funded startups, such as AI21, Cohere, Character.AI, and others, are racing with fresh capital to build larger algorithms in a bid to catch up with OpenAI’s current capabilities.
One of Cohere’s co-founders, Nick Frosst saw where Altman is coming from. He spoke with Wired and explained how progress isn’t only tied to scaling a model. “There are lots of ways of making transformers way, way better and more useful, and lots of them don’t involve adding parameters to the model,”. He went on to mention that new AI architectures, human-to-machine feedback, and designs are likely where new advancements will come from.
This is similar to what happened in naval warfare during and after the age of the battleship. Though during the first World War, a ship’s tonnage, gun size, etc. were metrics that normally would determine lethality. New advancements in other technologies, such as aircrafts, completely changed the nature of naval battle. Removing the constant race of big and bigger battleships. And replacing it with other advancements.
Since Sam Altman brought up size and scaling for models, he was asked in the Zoom call if GPT- 4‘s training cost $100 million dollars. To that, he simply replied, “It’s more than that.”. But as for now. It’s pretty much a given that OpenAI’s competitors still need to create their larger models in order to compete with GPT– 4.