Even though OpenAI o1-previewโ€™s advanced reasoning helps consumers, I expect that it will be even more useful for agentic reasoning and planning.

Most LLMs have been optimized for answering questions primarily to deliver a good consumer experience, and weโ€™ve been able to โ€œgraftโ€ them into complex agentic workflows to build valuable applications. The trend of LLMs built to support particular operations in agents natively will create a lot of lift for agentic performance. Iโ€™m confident that large agentic performance gains in this direction will be realized in the next few years.



OpenAl o1-preview์˜ ๊ณ ๊ธ‰ ์ถ”๋ก ์ด ์†Œ๋น„์ž์—๊ฒŒ ๋„์›€์ด ๋˜๋”๋ผ๋„, ์—์ด์ „ํŠธ ์ถ”๋ก  ๋ฐ ๊ณ„ํš์— ํ›จ์”ฌ ๋” ์œ ์šฉํ•  ๊ฒƒ์œผ๋กœ ๊ธฐ๋Œ€ํ•ฉ๋‹ˆ๋‹ค.

๋Œ€๋ถ€๋ถ„์˜ LLM์€ ์ฃผ๋กœ ์ข‹์€ ์†Œ๋น„์ž ๊ฒฝํ—˜์„ ์ œ๊ณตํ•˜๊ธฐ ์œ„ํ•ด ์งˆ๋ฌธ์— ๋‹ตํ•˜๋„๋ก ์ตœ์ ํ™”๋˜์—ˆ์œผ๋ฉฐ, ์šฐ๋ฆฌ๋Š” ์ด๋ฅผ ๋ณต์žกํ•œ ์—์ด์ „ํŠธ ์›Œํฌํ”Œ๋กœ์— "์ด์‹"ํ•˜์—ฌ ๊ฐ€์น˜ ์žˆ๋Š” ์• ํ”Œ๋ฆฌ์ผ€์ด์…˜์„ ๊ตฌ์ถ•ํ•  ์ˆ˜ ์žˆ์—ˆ์Šต๋‹ˆ๋‹ค. ์—์ด์ „ํŠธ์—์„œ ํŠน์ • ์ž‘์—…์„ ๊ธฐ๋ณธ์ ์œผ๋กœ ์ง€์›ํ•˜๋„๋ก ๊ตฌ์ถ•๋œ LLM์˜ ์ถ”์„ธ๋Š” ์—์ด์ „ํŠธ ์„ฑ๊ณผ์— ๋งŽ์€ ํ–ฅ์ƒ์„ ๊ฐ€์ ธ์˜ฌ ๊ฒƒ์ž…๋‹ˆ๋‹ค.

์ €๋Š” ์ด ๋ฐฉํ–ฅ์œผ๋กœ ํฐ ์—์ด์ „ํŠธ ์„ฑ๊ณผ ํ–ฅ์ƒ์ด ํ–ฅํ›„ ๋ช‡ ๋…„ ์•ˆ์— ์‹คํ˜„๋  ๊ฒƒ์ด๋ผ๊ณ  ํ™•์‹ ํ•ฉ๋‹ˆ๋‹ค.



https://www.deeplearning.ai/the-batch/issue-275/