The theme of this week is related to ChatGPT. In early March, OpenAI opened up the interface for ChatGPT. Various related applications have emerged like mushrooms after rain in the past two weeks.
Some of these products have become essential for me every day. At the same time, I am also thinking about how to use these tools to improve my learning and work efficiency. Especially when using chatPDF to read articles or papers, constantly asking questions helps me to focus more.
Products I have used#
Articles I have read#
I was doing a translation on prompt engineering some time ago. After reading this article, I finally understood why it is necessary to use prompts to control the output of ChatGPT.
Principle
"A generative model is a function that can take a structured collection of symbols as input and produce a related structured collection of symbols as output."
Deterministic VS Stochastic
Relationship matters
- Diversity and complexity of symbolic relationships
- Latent space -> multidimensional space of possible output results
Probability distributions
- Methods that can change the probability distribution of the LLM model
- Training
- Fine-tuning
- Reinforcement learning from human feedback (RLHF)
Some thoughts#
Technologies I have learned#
- paul-graham-gpt
- Using embedding to implement text search => Using cosSim to calculate similarity
export const cosSim = (A: number[], B: number[] ) =>
{
let dotproduct = 0
let MA = 0
let MB = 0
for (let I i < A.length; i++) {
dotproduct += A[i] * B[i]
MA += A[i] * A[i]
MB += B[i] * B[i]
}
MA = Math.sqrt(mA)
MB = Math.sqrt(mB)
const similarity = dotproduct / (MA * MB)
return similarity
}
- openai-translator
- Using tauri to package web applications into desktop apps
Opportunities for independent developers#
- The developer of typingmind earned $20,000 in one week