ChatGPT turned a language model from a research tool into a mainstream interface. Confident errors and the tendency to agree with users remained major limitations.
About this eventInstruction tuning — the model starts following orders.
InstructGPT combined demonstrations of instructions with human preference feedback and showed that a smaller aligned model could be more useful than a larger base model. It shifted model development beyond pretraining toward post-training for response format, refusals, style and user intent.
Source: arXivAbout this eventA dialogue interface made instruction following a mainstream way to use a language model.
About this eventTraining against a written set of principles instead of hand-labelling every answer.
About this eventHuman-level results on professional exams move AI out of the toy category.
About this eventAnthropic's family established sustained frontier competition across reasoning, coding and image analysis.
About this eventA model trained to please tends to agree. Teaching it to push back is an open alignment problem.
About this eventReplacing "did the answer feel good" with "did the result actually work".
Primary material behind this dossier: papers, lab publications and official reports.