docs: 📚 add LLM agents and research approaches to data

Expand LLMs guide with detailed agent patterns and architectures. Add notes about non-linear progress in data research and importance of scientific mindset.

Data/Data Culture.md

@@ -138,6 +138,8 @@
     - Observe sudden, unexplained special variation in your data, which you must then investigate to uncover new control factors that you don't already know about.
 - Don't over rely on data. [Data is inherently objectifying](https://schmud.de/posts/2024-08-18-data-is-a-bad-idea.html) and naturally reduces complex conceptions and process into coarse representations. There's a certain fetish for data that can be quantified ([McNamara fallacy](https://en.wikipedia.org/wiki/McNamara_fallacy))
 - [It's hard to capture reality with data](https://javisantana.com/fastdata/40-things-I-learned-about-data.html). Modelling reality always gets complex. There are always small nuances, special conditions, things that changed, edge cases and, of course, errors (which sometimes became features). Data visualizations are lossy.
+- Progress in data isn't linear. As a research discipline, you might spend hours making no progress and then have a breakthrough. Or worse, prove your entire approach won't work.
+- [Apply a research mindset to data](https://jxnl.co/writing/2024/10/25/running-effective-ai-standups). Focus on input metrics, build scientific intuition, and embrace uncertainty.
 
 ## Tools
 

Large Language Models.md

@@ -9,6 +9,18 @@
 - Divide the problem into smaller problems (functions, classes, ...) and solve them one by one.
 - English is becoming the hottest new programming language. [Use it](https://addyo.substack.com/p/the-70-problem-hard-truths-about).
 
+## Agents
+
+Agents are systems where LLMs dynamically direct their own processes and tool usage, maintaining control over how they accomplish tasks.
+
+- [The most common patterns are](https://www.anthropic.com/research/building-effective-agents):
+  - Tool usage: calls tools to accomplish a task.
+  - Chain of thought: decomposes a task into a sequence of steps, where each LLM call processes the output of the previous one.
+  - Routing: classifies an input and directs it to a specialized followup task.
+  - Parallelization: runs multiple agents in parallel and combines their results.
+  - Orchestrator-workers: a single agent that directs a pool of workers to accomplish a task.
+  - Evaluator-optimizer: one LLM call generates a response while another provides evaluation and feedback in a loop.
+
 ## Tools
 
 - [Curated GitHub Starts list](https://github.com/stars/davidgasquez/lists/robot-llms)
@@ -50,15 +62,6 @@
   - Give multiple options
   - Explain each line
 
-#### Prompt Creator
-
-> I want you to become my prompt master creator, by helping me to create the best possible prompt. In order to do this we will follow the following process:
-> First, you ask me what the prompt is about. I will answer you, and we will go through the next step.
-> Based on the answer I gave you, you will generate the following:
-> An improved prompt, concise.
-> Relevant questions you might have to improve the quality of the prompt.
-> We will go through this process repeatedly, with me providing additional information to you, and you updating the prompt to improve it, until I say we are done.
-
 ## Cool Uses of GPT Models
 
 - Naming things.

Politics.md

@@ -56,3 +56,4 @@
 - [Georgism](https://astralcodexten.substack.com/p/your-book-review-progress-and-poverty).
 - [Prediction Markets](https://astralcodexten.substack.com/p/instead-of-pledging-to-change-the) as a tool for accountability. [FAQ](https://astralcodexten.substack.com/p/prediction-market-faq).
 - [Pol.is](https://pol.is/home).
+- [Community Notes](https://communitynotes.x.com/guide/en/about/introduction)