How Large Language Models Evolved
And Why It Matters Today
Ever wondered how AI went from simple autocomplete to assistants that explain jokes or summarise books? That's the power of Large Language Models (LLMs). This guide walks you through their evolution-designed for learners, educators, and curious minds. Tap any model in the app's timeline to explore its moment in AI history.
🔗 Click the image below to enter the interactive app.
It's not just a diagram-it's your way to explore Large Language Models, moments, and metrics that shaped the AI we use today.
2018-2020: Foundations of Language Understanding
GPT-1 and BERT: Learning the Basics
- GPT-1 (2018): Introduced pre-training-learning from vast text collections before tackling specific tasks.
- BERT (2018): Read text in both directions, dramatically improving language comprehension.
💡 Jargon tip: A parameter is like a dial in the model's brain. More parameters usually mean more capacity to learn and respond intelligently.
GPT-2 and T5: Scaling Up and Unifying Tasks
- GPT-2 (2019): Could generate stories from a single prompt-so impressive, its release was initially withheld.
- T5 (2019): Framed every language task as “text in, text out,” unifying translation, summarisation, and more.
GPT-3 (2020): The Few-Shot Revolution
- GPT-3: At 175 billion parameters, it could perform tasks with just a few examples-no retraining required.
- It wrote essays, translated languages, and solved basic maths using natural instructions.
2021-2022: Smarter, Safer, More Conversational
- ChatGPT (2022): Based on GPT-3.5, it used human feedback to become more helpful and dialogue-friendly.
- LaMDA (Google) and PaLM: Focused on open-ended, safe conversation and deep reasoning.
- BLOOM and OPT: Offered powerful open models-available without a tech giant's budget.
2023: The Open-Source Boom
- LLaMA (Meta): Small enough to run locally, yet powerful enough for real applications.
- Claude (Anthropic): Followed a written “constitution” of ethical principles to guide responses.
- Mistral 7B and Yi-34B: Proved that smaller, well-trained models could outperform much larger ones.
Explore 2023 in the timeline to see which models were fully open-and how regions beyond the United States began shaping the future of LLMs.
2024-2025: Multimodal and Memory-Powered AI
- Gemini 1.5 (Google): Can handle 1 million tokens-enough to process entire books or documents.
- Claude 3 and GPT-4 Turbo: Added voice and image capabilities, making AI truly multimodal.
- Grok, Inflection-2, and DeepSeek-V2: Showed innovation is thriving beyond the biggest tech players.
Our Interactive Timeline App
What Can You Discover in the App?
Try exploring these questions as you move through the timeline:
- Which models gave the best performance for the least compute?
- Where did open-source models emerge-and who built them?
- What's the largest model with a context window over 100,000 tokens?
Use the filter tools in the app to explore by year, region, licence type, or benchmark performance.
Where Are We Now?
By 2025, Large Language Models do far more than generate text-they see, hear, reason, and hold real-time conversations. From tutoring students to analysing contracts, they now power tools across education, healthcare, law, entertainment, and beyond.
But it's not just the models that have evolved-so has how we explore them.
Our interactive app lets you:
- Search and filter by developer, region, or licence type
- Compare models using timeline charts on:
- Parameter growth
- Context window size
- MMLU benchmark scores
- Training compute costs
- Efficiency: compute cost per 1% MMLU gain
- Dive deep into detailed model summaries
Whether you're learning, teaching, or just curious-your journey through the history of AI starts here.
Want to Feel More Confident Using AI?
Understanding AI's history is just the start. With 1-on-1 coaching, we help you build real-world AI literacy-on the job, at your pace. Learn how to use tools effectively, write better prompts, and make sense of what's happening behind the scenes. You can even learn to build interactive visuals like the app you just explored.
Ready to get started? Let's talk.