Sign up to save tools and stay up to date with the latest in AI
bg
bg
1

Ask HN: Is it feasible to train my own LLM?

Jan 02, 2024 - news.ycombinator.com
The author, a high school student in the International Baccalaureate (IB) program, is planning to write a research paper on the impact of training data on Language Model's (LLM) output. They plan to compare the output of an LLM trained on Wikipedia data versus Reddit data. They have access to Nvidia GPUs for training and have a basic understanding of technology and programming.

However, they express concerns about their limited programming skills and the potential quality of their results. They are seeking feedback on whether this project is feasible and if the results, despite potentially being poor, would be sufficient to measure differences in the LLM's output based on the training data.

Key takeaways:

  • The author is an IB diploma candidate in high school planning to write a research paper on how an LLM's training data impacts its output.
  • The author plans to compare the output of an LLM trained on different sources, such as Wikipedia and Reddit.
  • The author has access to powerful Nvidia GPUs and plenty of time for training the LLM.
  • Despite having some technical skills, the author admits to having weak programming abilities and is seeking advice on the feasibility of the project.
View Full Article

Comments (0)

Be the first to comment!