# [TECH] AI & Large Language Models

**Artificial Intelligence** and **Large Language Models (LLMs)** are the current frontier of AI: neural networks trained on internet-scale data that can converse, reason, write code, generate images, and assist with scientific discovery.

## Overview

GPT-3 (OpenAI, 2020, 175B parameters) demonstrated that scaling language models produces qualitatively new capabilities. GPT-4 (2023), Claude (Anthropic, 2023), and Gemini (Google DeepMind, 2023) perform at human expert level across many domains. AlphaFold (2020) solved protein structure prediction; AlphaCode (2022) competes with professional programmers; AlphaGeometry (2024) solves olympiad geometry problems. AI is now being deployed in drug discovery, materials design, climate modelling, scientific literature synthesis, and industrial automation.

## Key Actors

- **Companies**: OpenAI (2015), Anthropic (2021), Google DeepMind (2023 merger), Meta AI, Mistral, Cohere, xAI
- **Investors**: Microsoft (USD 13B in OpenAI), Google (USD 400M in Anthropic), Amazon (USD 4B in Anthropic)

## Key Technologies

- Transformer architecture (Vaswani et al., 2017)
- Reinforcement learning from human feedback (RLHF)
- Constitutional AI (Anthropic)

## Economic Value

AI market: **USD 200 billion/year** (2023, Grand View Research). Goldman Sachs (2023) projects AI could add **USD 7 trillion/year** to global GDP within 10 years. McKinsey estimates USD 4.4T/year in value from generative AI alone by 2030.

## Notes

Goldman Sachs *The Potentially Large Effects of Artificial Intelligence on Economic Growth* (2023). McKinsey *The Economic Potential of Generative AI* (2023). Grand View Research *AI Market* 2023.

## What This Enables

This is a current frontier node — no downstream connections yet recorded in this graph.

## Discovery Character
⏎
**Surprise level**: Extreme — GPT-3's emergent abilities (2020) — few-shot learning, code generation, arithmetic — surprised OpenAI's own researchers. The capabilities of GPT-4 (2023) exceeded the predictions of most AI researchers by years. Scaling laws (performance improves predictably with compute and data) were discovered empirically; the emergent capabilities at each scale transition were not predicted from the laws.
⏎
**Mode**: Systematic with emergent surprises. Training large language models is systematic: scale up data, compute, and model size according to known scaling laws. But the capabilities that emerge at each scale threshold — in-context learning, chain-of-thought reasoning, code synthesis — were not predicted and surprised researchers. The combination of systematic training infrastructure and unpredicted emergent intelligence is the defining characteristic of the current AI moment.
⏎
# Parents

* [TECH] Cloud Computing & Big Data
* [SCI] Genomics & Computational Biology
* [TECH] Medical Imaging (X-ray, CT, PET)
* [TECH] Cloud Computing & Big Data
* [SCI] Deep Learning
* [TECH] Internet & World Wide Web