OpenAI has unveiled its latest model, GPT-4.5, which is designed to be more versatile than the company’s previous STEM-focused reasoning models. The model excels in tasks such as writing, programming, and solving practical problems. OpenAI CEO Sam Altman described GPT-4.5 as the first model that feels akin to conversing with a thoughtful individual. He expressed amazement at receiving genuinely good advice from the AI. However, Altman noted that the model is both ‘giant’ and ‘expensive’ to train and operate. Due to GPU shortages, the rollout has been limited to paid subscription tiers, with plans to expand access soon. OpenAI’s chief research officer, Mark Chen, highlighted that scaling continues to yield significant improvements without diminishing returns. Although training costs are high, inference costs have decreased substantially since the launch of GPT-4. GPT-4.5 is currently available to ChatGPT Pro users and developers via API, with broader access expected next week. The model represents an advancement in unsupervised learning, focusing on scaling up word knowledge, intuition, and reducing hallucinations. Compared to the o1 series, GPT-4.5 is better at picking up social cues and is less prone to hallucinations, making it ideal for various practical applications. — news from Business Insider
