https://paper.kim/en/categories/ai/ Recent content in Ai on paper.archive https://paper.kim/images/default-thumb.jpg https://paper.kim/images/default-thumb.jpg Hugo en-US Sun, 06 Sep 2026 22:00:00 +0900 https://paper.kim/en/posts/ai/2026-09-06-gpt6-astra-model-comparison/ Sun, 06 Sep 2026 22:00:00 +0900 https://paper.kim/en/posts/ai/2026-09-06-gpt6-astra-model-comparison/ GPT-6 Astra shipped on September 3. Compared against Claude Fable 5.1, Opus 5 and GPT-5.6 Sol: the 2.5x price increase, the token-efficiency twist, the ARC-AGI-3 harness problem, and the row missing from the announcement. https://paper.kim/en/posts/ai/2026-09-01-fable51-model-comparison/ Tue, 01 Sep 2026 21:00:00 +0900 https://paper.kim/en/posts/ai/2026-09-01-fable51-model-comparison/ Claude Fable 5.1 shipped today. Compared against Opus 5, Fable 5 and GPT-5.6 Sol: the 75% cache-read price cut, three breaking changes, and where the vendor table and third-party measurements disagree. https://paper.kim/en/posts/ai/2026-07-28-llm-wiki/ Tue, 28 Jul 2026 00:00:00 +0900 https://paper.kim/en/posts/ai/2026-07-28-llm-wiki/ A walkthrough of Karpathy's LLM Wiki pattern: the three-layer structure, the ingest/query/lint loop, how people actually use it, and how much of the efficiency claim has actually been measured. https://paper.kim/en/posts/ai/2026-07-25-sf-ai-declaration/ Sat, 25 Jul 2026 09:00:00 +0900 https://paper.kim/en/posts/ai/2026-07-25-sf-ai-declaration/ What the San Francisco AI Declaration of July 24, 2026 actually says, which deals were signed alongside it, and what Korea's AI Framework Act requires from developers today. https://paper.kim/en/posts/ai/2026-07-25-opus5-model-comparison/ Sat, 25 Jul 2026 08:30:00 +0900 https://paper.kim/en/posts/ai/2026-07-25-opus5-model-comparison/ Claude Opus 5 compared with GPT-5.6 Sol, Fable 5 and Sonnet 5 across coding, agentic work, long context and cost, plus what changed from Opus 4.8 and what to watch when migrating.