Google’s NotebookLM artificial ielligence tool has recely attracted a lot of atteion from users; You can give your text files to this artificial ielligence to produce a podcast with its coes. Now Meta has unveiled NotebookLlama, which does almost the same thing but is open source.
NotebookLlama first makes a copy of your file, which could be a news article, TechCrunch reports. It then dramatizes that transcript, meaning, for example, that it inserts pauses in the text to make the final output sound more like a podcast. It then feeds the dramatized transcript version to text-to-speech models. You can see this process in the image below:
Meta artificial ielligence to compete with Google
What Meta AI does is similar to Google’s NotebookLM, but it doesn’t seem to have the same quality. In the examples published by NotebookLlama, we can see that the voices are robotic and sometimes the seences are not understandable. Meta researchers say the quality of podcasts could be improved with more robust models. In this project, Meta uses its own Llama models for processing.
Meta on the GitHub page of the NotebookLlama model says that one possible approach to creating a podcast is to use 2 models to discuss a common topic and write the outline of the podcast together. Of course, currely in this artificial ielligence, a single model is used to write the outline of the podcast.
Overall, NotebookLlama isn’t the first AI to attempt to emulate NotebookLM. Previously, some projects are more successful than competitors, but none of them – even NotebookLM itself – have been able to solve the problem of artificial ielligence illusion; That is, the podcasts produced by artificial ielligence sometimes coain fake information.
