Revision History
Causal Language Model · 8 revisions
Sizes are character counts of the article source. The signed number is the change from the previous revision.
Recent edit summaries
Detailed summaries recorded by editors. Generic maintenance summaries are omitted here; every recorded revision remains below. Dates describe the edit, not necessarily the event it covers.
- Correction: Corrected the KV-cache memory example (320 KB per token holds only with 8 grouped key-value heads), the GPU-hours needed to train a 100B-class model (about a million, not tens of thousands), the attribution of RLHF for language models (Ziegler 2019, not Christiano 2017) and of "think step by step" prompting (Kojima 2022, not Wei 2022), GPT-5 being a routed system rather than a single model, GPT-4o image generation arriving in March 2025, and brought the GPT lineage up to GPT-6 Astra (plus 13 minor fixes)
Version 8 · Nov 5, 2026, 01:53 AM
- Improve internal links (2026-10-13)
Version 7 · Oct 13, 2026, 10:01 AM
- Added 6 contextual internal links
Version 6 · Jul 23, 2026, 03:00 PM