OpenAI quietly released a new model in the middle of the night, introducing-openai-o1-preview. They released a series of videos showcasing the power of the new model, and the internet is flooded with articles discussing how impressive the new model’s evaluations are. However, having seen plenty of hype in the AI world, I approached it with a skeptical attitude and immediately tried it out firsthand.

Chinese Interpretations

Recently, Li Jigang had a very popular prompt that can generate interesting new interpretations of Chinese characters. I tried it with Claude3.5 and the results were particularly good. Below are some SVG images generated by Claude:

In LevelDB, the data in the memory MemTable is stored in a SkipList to support fast insertion. Skip lists are probabilistic data structures proposed by William Pugh in his paper Skip Lists: A Probabilistic Alternative to Balanced Trees. They are somewhat similar to ordered linked lists but can have multiple levels, trading space for time, allowing for fast query, insertion, and deletion operations with an average time complexity of $ O(\log n) $. Compared to some balanced trees, the code implementation is relatively simple and performance is stable, making it widely applicable.

So, what are the principles behind skip lists? How are they implemented in LevelDB? What are the highlights and optimizations in LevelDB’s skip list implementation? How does it support single-threaded writing and concurrent reading of skip lists? This article will delve into these aspects from the principles and implementation of skip lists. Finally, it provides a visualization page that intuitively shows the construction and overall structure of skip lists.

Recently, Anthropic released the system prompts for the Claude3.5 model, which are very worth learning from. The entire prompt is written in English, quite lengthy, and constrains many behaviors of the model. Let’s take a look together.