However, Dirk Hohndel, the interviewer, expresses concerns about the potential for LLMs to "hallucinate" or make up stuff, which could lead to bugs if there isn't a human to catch them. Torvalds counters this by pointing out that bugs occur even without the use of LLMs. The discussion also references a previous statement by Torvalds in 2015 about the development of AI through recurrent neural networks, and an article by Karpathy on the effectiveness of these networks.
Key takeaways:
- Linus Torvalds, the creator of Linux, believes that large language models (LLMs) will be used to write code and possibly already are on a smaller scale.
- Torvalds hopes that LLMs will be able to assist in code reviews and maintenance by identifying bugs and irregular patterns.
- Despite the potential for LLMs to 'hallucinate' or make errors, Torvalds remains optimistic about their use, arguing that humans already make plenty of errors without them.
- Torvalds' views on AI have been consistent since 2015, when he predicted the rise of AI through something like recurrent neural networks, which would not be 'reliable' in the traditional sense due to the need for training.