C.W.K.
Stream
Lesson 06 of 10 · published

Impact: From Translation to GPT, BERT, and the Modern LLM Stack

~16 min · history, gpt, bert, llm

Level 0Token
0 XP0/94 lessons0/10 achievements
0/120 XP to next level120 XP to go0% complete

The Transformer was published in 2017 as a translation model. Within two years it had escaped translation and become the universal substrate for almost every NLP system that mattered. Within five years it had escaped language entirely.

The pivotal early follow-ups:

  • GPT-1 (June 2018, OpenAI): a 117M-parameter decoder-only Transformer pre-trained on next-token prediction over BooksCorpus, then fine-tuned for downstream tasks. The "pre-train then fine-tune" recipe is born.
  • BERT (October 2018, Google): a 340M-parameter encoder-only Transformer pre-trained with masked language modeling. State of the art on every NLU benchmark for years.
  • T5 (October 2019, Google): "every NLP task is text-to-text." Uses the full encoder-decoder shape and unifies translation, summarization, classification, and QA.
  • GPT-3 (May 2020, OpenAI): 175B parameters, demonstrating that scale unlocks in-context learning — solving new tasks just from examples in the prompt, no fine-tuning required.

By 2024-2026, every frontier model — GPT-4/5, LLaMA 4, Gemini 2.5 Pro, Claude 3.7 Sonnet, Mistral Large, Qwen 3 — is a Transformer variant. The unit cell from the 2017 paper is unchanged; only its scale, training data, and post-training (RLHF, DPO, GRPO, Constitutional AI) have evolved.

Code

Family tree (rough)·text
2017  Transformer (encoder-decoder, translation)
  ├── 2018  GPT-1     (decoder-only, next-token)        ─┐
  ├── 2018  BERT      (encoder-only, masked LM)          │
  ├── 2019  GPT-2 / RoBERTa / T5 / BART / XLNet         │
  ├── 2020  GPT-3 (175B, in-context learning)            ├─→ all decoder-only
  ├── 2022  Chinchilla / PaLM / Codex                    │   from 2022
  ├── 2023  LLaMA / GPT-4 / Claude 2 / Mistral 7B        │
  ├── 2024  LLaMA 3.x / Gemini 1.5 / Claude 3 / Mixtral  │
  └── 2025  GPT-5 / LLaMA 4 / Gemini 2.5 / Claude 3.7   ─┘

External links

Exercise

Pick three modern frontier models (e.g., GPT-5, LLaMA 4 Scout, Gemini 2.5 Pro) and locate the official architecture details (parameter count, layers, d_model, attention type, vocab size, context length) in their model cards or release notes. Tabulate them. Notice which numbers are public vs. unstated, and which design choices are now nearly universal.

Progress

Progress is local-only — sign in to sync across devices.
Spotted a bug or have feedback on this page?Report an Issue

Comments 0

🔔 Reply notifications (sign in)
Sign inPlease sign in to comment.

No comments yet — be the first.