Actualités

Written By: Chloe Zheng

According to Sequoia Capital’s 2023 research, 85% of developers favor fine-tuning existing models over training from scratch. Recent development underscores this trend: DeepSeek has open-sourced its models and implemented model distillation, a process where larger “teacher” models transfer inference logic to smaller “student” models, optimizing knowledge compression and performance. Similarly, OpenAI’s ChatGPT o3 release further emphasizes post-training and reinforcement learning, further highlighting the industry’s focus on refining existing AI capabilities.