November 2023 Summaries

Kyle Corbitt analyzed data from Hacker News comments on crypto and AI to look for similarities and differences between the two. He found that while sentiment towards AI has always been higher than sentiment towards crypto, AI sentiment steadily dropped from 2010-2018 and has been hovering around neutral for the last 5 years. Crypto sentiment has a much weaker correlation with the crypto hype cycle than expected. The analysis also showed that fine-tuning a model on a specific dataset can improve accuracy and reduce costs compared to using general-purpose prompted models like GPT-3.5 or GPT-4, which cost $1K and $10K respectively.

Mistral, a new model developed by Meta AI, has been found to outperform similar small models on benchmarks and offers an 8K context window, making it suitable for fine-tuning task-specific LLMs under 34B parameters. The performance improvements of Mistral over Llama 2 are significant and generalize across many task types, with users consistently preferring Mistral's output or considering it comparable to Llama 2 13B in non-deterministic tasks. While there are areas yet to be evaluated, such as multilingual tasks, overall, Mistral is an extremely strong model that can save customers significant money by decreasing inference costs while improving accuracy and reducing latency.