OpenAI has introduced GPT-4.5, and it truly is a step forward.
What’s interesting:
— This is the first model that feels like a conversation with a thoughtful person. Some users note that they received really useful advice from the AI.
— The model is not geared towards winning benchmarks — GPT-4.5 is not about logic, but about a new type of intelligence that feels different.
What’s not right:
— GPT-4.5 is a huge and expensive model, and OpenAI currently lacks GPUs for all users. It is available in Pro, and it will be added to Plus after a large-scale server expansion.
OpenAI continues to scale — tens of thousands of new GPUs will appear next week, and hundreds of thousands in the near future.
🔗 Source: https://youtu.be/cfRYp0nItZ8
P.S. Andrej Karpathy, who has been testing the model for the last couple of days, notes that each 0.5 version on average improves quality by 20%, but requires 10 times more computational effort during training. The result: again the feeling - it has become cooler, but with each subsequent step, it gets harder and harder. I think the creation of AGI is postponed for a while (a breakthrough in algorithms or in computing power).
