AIAcademy update cohort · archived

2022 Q4 pre-ChatGPT foundation stack

The archived 2022 Q4 baseline preserves the pre-ChatGPT foundation stack: transformer architecture, GPT-3 few-shot prompting, instruction tuning, compute-optimal scaling, ReAct-style tool reasoning, and early multimodal research before consumer assistants made those ideas visible to everyday learners.

5 updates · 5 source-linked · 0 product-created

Key questions

• Which pre-ChatGPT research ideas became the durable foundation for later assistant lessons?
• Which 2022 claims should learners treat as architecture, training, and evaluation context rather than current product advice?
• Which early tool-use and multimodal papers became the baseline for later agent and media-literacy guidance?

Updates

• Transformer architecture baseline preserved
  The 2022 archive keeps transformer architecture visible as the pre-ChatGPT substrate for later lessons about scale, instruction tuning, agents, and multimodal systems.
• Pretraining and scaling baseline archived
  GPT-3, PaLM, and Chinchilla stay visible as the historical baseline for explaining pretraining, few-shot behavior, and why data-compute balance matters before assistant tuning.
• Instruction and alignment baseline archived
  InstructGPT, Sparrow, and Constitutional AI remain archived as the pre-ChatGPT alignment baseline before later cohorts add system cards, safety policy, and product guardrails.
• Reason-act tool-use baseline archived
  ReAct stays visible as the 2022 baseline for reasoning-plus-action patterns before product tool calling, API schemas, and agent deployment controls became everyday builder concerns.
• Early multimodal research baseline archived
  Flamingo remains archived as the early visual-language baseline before Gemini, GPT-4o, and current video-generation products changed what multimodal literacy needs to cover.