×
Please Note! This page is still a work in progress.
Large Languagle Models (LLMs) & Natural Language Processing (NLP)
Est. read time: 1 minute | Last updated: March 31, 2025 by John Gentile
Contents
Overview
Large Language Models (LLM)
Prompt Engineering
Suggestions:
- Start with a short and simple prompt, and iterate to get better results.
- Put instructions at beginning or end of prompt, while clearly separating instructions from the text of interest.
- Describe how the model should behave and respond- for example, if looking for coding advice, can create a system prompt of
You are a helpful assistant that answers programming questions.
- Add specificity and descriptions about the task and desired output, as well as including examples of output if possible.
- Instructions should tell what “to do” rather than “what not to do”.
References:
LLMs for Coding
- (3/27/2025) so far Claude 3.7 Sonnet has slightly better coding results.
LLMs for Research
Claude, OpenAI/ChatGPT and Grok have “research” modes where the model is open to searching the internet, and has multi-step reasoning to come up with deep assessments and recommendations.
Self-Hosted
- LLaMA: Open and Efficient Foundation Language Models - arXiv: Meta AI open-source LLM model.
- llama3 implemented from scratch
- llama.cpp: fast, low overhead inference of LLaMA in C/C++.
Other Uses/Automation
- LaurieWired/GhidraMCP: ghidraMCP is an Model Context Protocol server for allowing LLMs to autonomously reverse engineer applications. It exposes numerous tools from core Ghidra functionality to MCP clients.
References
- Attention Is All You Need- Arxiv: introduces concepts of transformers and attention
- Hugging Face: pre-trained NLP models & reference
- Training data-efficient image transformers & distillation through attention- Facebook AI
- The Illustrated Transform: NLP walk-through