OpenAI released GPT-4.5 on Thursday, marking it as the largest and most powerful model to date. CEO Sam Altman characterized it as a breakthrough, claiming it provides advice and responses that feel genuinely thoughtful. The company aims to expand accessibility to the model, initially offering it to users of the ChatGPT Pro API and planning to roll it out to more subscription tiers, citing limitations in GPU resources.
GPT-4.5 focuses on enhancing data processing and computational strategies, maintaining that increased scale results in improved performance. Despite high training costs, OpenAI has successfully reduced inference costs significantly. The model is designed with advancement in “unsupervised learning,” emphasizing improved social interaction skills and emotional intelligence, enabling it to better understand and respond to social cues compared to previous models like O1.
In a live demonstration, staff highlighted GPT-4.5’s ability to reconstruct sensitive texts with greater tact and warmth, suggesting it is a better conversational partner. The model claims to be more accurate with reduced hallucinations, making it suitable for various tasks, from writing to problem-solving. Overall, GPT-4.5 represents a significant evolution in AI capabilities, emphasizing emotional understanding and user alignment.
Source link
