@defoemark Awesome work. Could you tell more about the requriments for the training data? 30 sec single voice snippets with text? Or will shorter snippets work?
@GoogleDeepMind@ashkamath20@GoogleDeepMind Sounds like a fantastic starting point for adding more languages. Any chance that you also can release a small dataset to prevent forgetting current knowledge?
@MTrucksa@Math_files The easiest way to picture it is probably to picture what is left. 1/6 is smaller than 1/5. Therefore 5/6 has taken the largest chunk of the pizza. However, as the clever kid pointed out, nothing indicates that the pizzas has the same size.
@kenneth0stanley@hardmaru Understanding (comprehension) and compression is fundamentally the same prosess. There are multiple ways to understand/compress information. Ideally both should be optimised toward future utility.
@thegenioo@OpenAI I feel 4.5 is quite a lot better at writing. Especially more tricky writing styles. Qualitatively. It just feels better, and does not look AI generated the same way as 4o does. For programming I have had more luck with 4o.
@gunsnrosesgirl3 No. This is about focal distance of the lens used in the example. Try the following experiment: In the section where the train is moving fast. Cut out a small frame in a piece of paper. Hold over your screen to block the peripherals. The train still moves fast!