The Era of Large AI Models Is Over

Just kidding: we all know size matters. It’s definitely true for AI models, especially for those trained on text data, i.e., language models (LMs). If there’s one trend that has, above all others, dominated AI in the last five or six years, it is the steady increase in parameter count of the best models, which I’ve seen referred to as <a href="https://huggingface.co/blog/large-language-models" rel="noopener ugc nofollow" target="_blank">Moore’s law for large LMs</a>. The GPT family is the clearest — albeit not the only — embodiment of this fact: GPT-2 was 1.5 billion parameters, GPT-3 was 175 billion, ~100x its predecessor, and rumors have it that GPT-4’s size, although officially undisclosed, has reached the <a href="https://www.semafor.com/article/03/24/2023/the-secret-history-of-elon-musk-sam-altman-and-openai" rel="noopener ugc nofollow" target="_blank">1 trillion mark</a>. Not an exponential curve but definitely a growing one. OpenAI was categorically following the godsend guidance of the scaling laws they themselves <a href="https://arxiv.org/abs/2001.08361" rel="noopener ugc nofollow" target="_blank">discovered</a> in 2020 (that DeepMind later <a href="https://arxiv.org/abs/2203.15556" rel="noopener ugc nofollow" target="_blank">refined</a> in 2022). The main takeaway is that size matters a lot. DeepMind revealed that other variables like the amount of training data, or its quality, also influence performance. But a truth we can’t deny is that we love nothing more than a bigger thing: Model size has been the gold standard for heuristically measuring how good an AI system would be. <a href="https://medium.com/@albertoromgar/the-era-of-large-ai-models-is-over-a5c9d7d804d4">Read More</a>