The Minutes of Dylan Patel on o1 https://minutesof.com/dylan-patel/on/o1/ Everything Dylan Patel has said on o1: 5 verbatim quotes between November 2024 and December 2024, each with a timestamp and a link to the recording it… en Sun, 30 Aug 2026 16:22:19 +0000 Patel reveals o1's reasoning process sometimes switches between Chinese and English during hidden thinking pha https://minutesof.com/q/1524c47b-e0e1-402a-b743-111da2d0e959/ https://minutesof.com/q/1524c47b-e0e1-402a-b743-111da2d0e959/ “It generates tons of things. It's like it it sometimes switches between Chinese and English. Right? Like, whatever it is. It's thinking. Right?” — BG2 Pod Mon, 23 Dec 2024 20:12:56 +0000 o1 Patel states OpenAI's o1 reasoning model increases average sequence length because it reasons before outputtin https://minutesof.com/q/95d406f2-b33e-4dfc-9526-6aa4bec2193d/ https://minutesof.com/q/95d406f2-b33e-4dfc-9526-6aa4bec2193d/ “This is really important because with reasoning models like OpenAI's o one, the average sequence length grows a lot.” — Scaling Intelligence Tue, 12 Nov 2024 04:17:00 +0000 o1 Patel says o1 generates 40k sequence lengths versus 4k for standard models, requiring lower batching and highe https://minutesof.com/q/2548d082-f15b-4fca-b6d4-4cfc177ee7e5/ https://minutesof.com/q/2548d082-f15b-4fca-b6d4-4cfc177ee7e5/ “If you ask it to, like, generate a web scraper, in the standard one it'll just start outputting code, and it'll be maybe like a four k sequence length.” — Scaling Intelligence Tue, 12 Nov 2024 04:17:00 +0000 o1 Patel explains OpenAI's o1 model thinks for 5-20 seconds before outputting, creating inference throughput chal https://minutesof.com/q/179104b9-35d2-44c2-9d8a-85d97e3bd8f8/ https://minutesof.com/q/179104b9-35d2-44c2-9d8a-85d97e3bd8f8/ “When you use OpenAI's o one, it thinks for ten seconds, twenty seconds, five seconds. It varies a lot, but thinks for a while and then it sends you tokens.” — Scaling Intelligence Tue, 12 Nov 2024 04:17:00 +0000 o1 Patel explains o1's thinking time creates memory bandwidth issues that prevent batching users at high levels. https://minutesof.com/q/b7e7fd8f-2e86-4705-b53e-c249725aab84/ https://minutesof.com/q/b7e7fd8f-2e86-4705-b53e-c249725aab84/ “But if you batch higher, k b cache is not just a memory capacity issue, it's also a memory bandwidth issue.” — Scaling Intelligence Tue, 12 Nov 2024 04:17:00 +0000 o1