Skip to content
PromptsBuddy
All episodes
Season 1 · Episode 09September 5, 202625 min

Podcast episode

The Lineage — Episode 01: The Papers That Taught Machines to Read

The PromptsBuddy Podcast

2000–2026: from phrase tables and n-grams to transformers, GPT-3, InstructGPT, open weights, and what’s inside a September 2026 checkpoint. Rony and Meghna walk the primary papers — no demos, no skipped boring years.

  • MeghnaStaff engineer

The PromptsBuddy Podcast

The Lineage — Episode 01: The Papers That Taught Machines to Read

Year 2000 — counting

0:000:00

Chapters

  1. 00:00Year 2000 — counting
  2. 00:45The Lineage
  3. 01:11SMT, BLEU, CRF, Bengio
  4. 04:23Word2Vec to attention
  5. 07:11Attention Is All You Need
  6. 08:09ELMo, GPT-1, BERT
  7. 10:12Scaling, GPT-3, CLIP
  8. 12:55InstructGPT and the text box
  9. 14:22Break — three papers
  10. 14:50RoPE, GQA, retrievers
  11. 17:14Agents and swarms
  12. 19:08Open weights to 2026
  13. 23:28Reading list / next week

Show notes

The Lineage, episode one of three. Text and multimodal understanding from 2000 to September 2026.

Rony asks what changed. Meghna reads the primary papers — including the ones that sit in a 2026 checkpoint and never got a conference talk. No employer affiliations. Paper facts only.

Next week: image and video generation. Then audio.

What we cover

  • 2000–2013 — SMT, BLEU, Bengio 2003, Word2Vec / GloVe / FastText
  • 2014–2017 — seq2seq, Bahdanau attention, GNMT, then Vaswani
  • 2018–2020 — ELMo, GPT-1, BERT, GPT-3, RAG, CLIP
  • 2022– — InstructGPT, ReAct, FlashAttention, open weights
  • What’s inside the file — RoPE, GQA, MLA, speculative decoding, retrievers
  • Agents to 2026 — Generative Agents → AutoGen → Kimi Agent Swarm
  • Still wrong — next-token ≠ true; long-context wobble; visual misses

Reading list (short)

Papineni 2002 · Koehn 2003 · Bengio 2003 · Vaswani 2017 · Brown 2020 · Ouyang 2022 · DeepSeek-R1 2025 · Su 2021 RoPE · Ainslie 2023 GQA · Leviathan 2023 · Reimers 2019 · Park 2023 · AutoGen · Kimi K2 report

Mentioned in this episode

Prompts from this episode

Researchintermediate

Read a paper without losing the caveats

Extracts what a study actually established, at what sample size, under what conditions — and separates the authors' claims from what the press release said.

ClaudeChatGPTPerplexity2 vars
Open prompt
Engineeringintermediate

Explain this codebase like I start on Monday

Produces the orientation doc a new engineer actually needs: the three things that will confuse them, the load-bearing files, and the conventions nobody wrote down.

ClaudeChatGPT3 vars
Open prompt
FundamentalsbeginnerStaff pick

Make it ask before it answers

Four lines that stop a model guessing at what you meant. It interviews you until it can actually do the task, which turns a mediocre first answer into a usable one.

ClaudeChatGPTGemini1 vars
Open prompt