The analysis reveals that Machine Learning (ML) consistently occupied a larger fraction of front-page stories over the last 13 years, with a peak in 2018. The sentiment towards AI has been higher than crypto, but has been neutral for the last five years. The sentiment towards crypto has a weaker correlation with the crypto hype cycle. The author concludes that modern Language Learning Models (LLMs) can solve problems in a few hours that would have taken an ML team weeks.
Key takeaways:
- The author used OpenPipe to analyze data from Hacker News posts and comments, focusing on the trends and sentiments towards AI and crypto.
- Instead of using naive string matching, the author used GPT-3.5 to classify the data, which was more cost-effective and efficient.
- The author also fine-tuned a Mistral 7B model on the dataset to improve accuracy, demonstrating the benefits of fine-tuning over general-purpose prompted models.
- The analysis revealed some interesting trends, such as a steady drop in sentiment towards AI from 2010-2018, and a weaker correlation between crypto sentiment and the crypto hype cycle than expected.