PANews reported on November 12 that according to Decrypt, Thomas Wolf, co-founder and chief scientific officer of AI startup Hugging Face, said at the Web Summit that the company is developing small language models (small LMs) to promote the development of "next-stage robotics" technology. Such small models can perform tasks under low-latency conditions, allowing robots to be used in real environments outside of production lines. Wolf believes that such models can be embedded in smartphones, laptops, and even home appliances to achieve device-localized AI processing.
Hugging Face released the SmolLM small language model this year, demonstrating the superiority of small models in tasks such as data processing, image processing, and speech recognition. Wolf pointed out that small models can achieve performance similar to large models on specific tasks, but with the advantages of speed and efficiency. In addition, Wolf predicts that AI will be divided into two major trends in the future: on the one hand, the ever-expanding frontier large models are used for scientific exploration, and on the other hand, small models are widely embedded in various devices, just like the Internet.