Large language models (LLMs)

• Natural Language Processing
  • Biomedical natural language processing
• Natural language understanding
• Capacities
  • Capacity
  • Similarity to humans - see also Personas
• Domain-specific
  • Language models for scholarly texts
  • Language models for health records
  • Language models for genetics
• Impact
  • Impact of LLMs on teaching
• Models
  • Open source models
• Training
  • Finetuning
    • Reinforcement learning from human feedback
    • LoRA
  • Synthetic data
    • Use of synthetic data
• Persona
  • Personas
• Visualization
  • Visualization

Overview

Language model implemented with huge Neural networks.

Neural language models perform the language modeling tasks by using a neural network rather than explicitly estimating the conditional probabilities, by learning good vector-space representations for words, sentences, etc.

Issues

Large language models are often called Stochastic parrot, given that they are simply models that predict the next (or missing) token one by one. One exemplary problem from this is that large language models often confidently lie.

They can suffer from Training data leakage in language models.

Topics

• Prompt tuning
• Search engine

Models

• Transformer model
  • GPT 4
  • Dolly
  • OLMo - a completely open model

Tutorials and reviews

• Intro to Large Language Models by Andrej Karpathy
• Understanding Large Language Models by Sebastian Raschka
  • Why the Original Transformer Figure Is Wrong, and Some Other Interesting Historical Tidbits About LLMs
• Large language models, explained with a minimum of math and jargon
• Anti-hype LLM reading list

From scratch

• https://github.com/rasbt/LLMs-from-scratch by Sebastian Raschka
• Let’s build GPT: from scratch, in code, spelled out by Andrej Karpathy
• Create a Large Language Model from Scratch with Python free code camp

Text generation

• Controllable Neural Text Generation by Lilian Weng

Memes