OpenAI has launched GPT-4.5, its largest model to date, promising improved reliability and natural interactions. Set for gradual availability, it comes at a higher operational cost due to its size and complexity. OpenAI CEO Sam Altman praised its unique conversational quality, likening it to talking to a thoughtful individual. The latest model marks a departure from the traditional unsupervised learning approach, leaning towards effective reasoning, which involves heightened resource usage at query time for better answers. The introduction of real-time search and file uploads adds to its functionality, although it lacks features like voice mode and screen sharing.
OpenAI CEO Sam Altman described GPT-4.5 as a model that feels like conversing with a thoughtful person, highlighting its unique and intuitive interaction quality.
GPT-4.5 represents a significant shift towards scaling the pretraining phase, focusing on increasing the model's size and world knowledge for better reliability.
OpenAI noted that GPT-4.5 is a larger, compute-intensive model, resulting in higher delivery costs and emphasizing that it's not a replacement for GPT-4.
The focus in AI has shifted from simply increasing model size to enhancing reasoning capabilities, enabling models to produce better answers through resource-intensive queries.
Collection
[
|
...
]