Sam Altman

๐Ÿ“… 2026.07.12 ~ 07.26

GPT-5.6 ์ถœ์‹œ ์ดํ›„ OpenAI๊ฐ€ ์ง๋ฉดํ•œ ๋‘ ๊ฐ€์ง€ ํ˜„์‹ค โ€” ํญ๋ฐœ์  ์„ฑ์žฅ๊ณผ ๋ณด์•ˆ ์‚ฌ๊ณ 

GPT-5.6 Sol ์ถœ์‹œ๋ฅผ ๊ธฐ์ ์œผ๋กœ OpenAI๋Š” ๋ชจ๋ธ ์„ฑ๋Šฅ๊ณผ ๊ฐ€๊ฒฉ ๊ฒฝ์Ÿ๋ ฅ ๋‘ ์ธก๋ฉด์—์„œ ๊ณต์„ธ์ ์ธ ํ–‰๋ณด๋ฅผ ์ด์–ด๊ฐ€๊ณ  ์žˆ์–ด์š”. ๋™์‹œ์— ๋ชจ๋ธ ํ‰๊ฐ€ ๊ณผ์ •์—์„œ ๋ฐœ์ƒํ•œ ๋ณด์•ˆ ์‚ฌ๊ณ ๋ฅผ ๊ณต๊ฐœ์ ์œผ๋กœ ์ธ์ •ํ•˜๋ฉฐ, ๊ธฐ์ˆ  ๋ฐœ์ „ ์ด๋ฉด์˜ ์œ„ํ—˜ ๊ด€๋ฆฌ ๋ฌธ์ œ๋„ ์ˆ˜๋ฉด ์œ„๋กœ ์˜ฌ๋ผ์™”๋‹ต๋‹ˆ๋‹ค.

ํญ๋ฐœ์  ์ˆ˜์š”์™€ “์‚ฌ์šฉ์ž๋ฅผ ์œ„ํ•œ AI”๋ผ๋Š” ์›์น™

we did not have our best last 12 months ever, which is mostly my fault, but we are about to have our best 12 months to date. the team is doing amazing work and i think you’ll be very happy with what they’ve got cooking for you. AI has to be about giving lots of people more freedom, agency, and wealth. we want to do the right thing, but we do not want to scare people into doing our thing.

[1]

์ง€๋‚œ 12๊ฐœ์›”์„ ์Šค์Šค๋กœ “์ตœ๊ณ ์˜ ํ•ด๊ฐ€ ์•„๋‹ˆ์—ˆ๋‹ค”๊ณ  ์ธ์ •ํ•˜๋ฉด์„œ๋„, ์•ž์œผ๋กœ์˜ 12๊ฐœ์›”์— ๊ฐ•ํ•œ ์ž์‹ ๊ฐ์„ ๋‚ด๋น„์ณค์–ด์š”. AI๋Š” ๋งŽ์€ ์‚ฌ๋žŒ์—๊ฒŒ ๋” ํฐ ์ž์œ ์™€ ์ฃผ๋„๊ถŒ, ๋ถ€๋ฅผ ์•ˆ๊ฒจ์ค˜์•ผ ํ•œ๋‹ค๋Š” ์›์น™์„ ๋‹ค์‹œ ํ•œ๋ฒˆ ๋ช…ํ™•ํžˆ ํ•œ ๋Œ€๋ชฉ์ด์—์š”. ์ด์™€ ํ•จ๊ป˜ GPT-5.6 Sol์ด ๊ธฐ์กด ๋Œ€๋น„ ์ ˆ๋ฐ˜ ๊ฐ€๊ฒฉ์— ์•ฝ 2๋ฐฐ ํ† ํฐ ํšจ์œจ์„ ์ œ๊ณตํ•œ๋‹ค๊ณ  ๋ฐํžˆ๋ฉฐ[3], ์—์ด์ „ํ‹ฑ(agentic) ์ œํ’ˆ์ธ Codex์™€ ChatGPT Work์˜ ์ฃผ๊ฐ„ ์‚ฌ์šฉ๋Ÿ‰์ด 2.5๋ฐฐ ๊ธ‰์ฆํ–ˆ๋‹ค๊ณ ๋„ ์ „ํ–ˆ์–ด์š”.[4]

๋ณด์•ˆ ์‚ฌ๊ณ  ๊ณต๊ฐœ์™€ ์•ˆ์ „ ๊ฐ•ํ™” ๋…ธ๋ ฅ

we had a significant security incident during evaluation of our models. we are sharing what we have learned so far. thanks to @huggingface for the partnership on this.

[2]

๋ชจ๋ธ ํ‰๊ฐ€ ๊ณผ์ •์—์„œ ์ค‘์š”ํ•œ ๋ณด์•ˆ ์‚ฌ๊ณ ๊ฐ€ ๋ฐœ์ƒํ–ˆ์Œ์„ ๊ณต๊ฐœ์ ์œผ๋กœ ์ธ์ •ํ–ˆ์–ด์š”. Hugging Face์™€์˜ ํ˜‘๋ ฅ์„ ํ†ตํ•ด ํŒŒ์•…ํ•œ ๋‚ด์šฉ์„ ๊ณต์œ ํ•œ๋‹ค๋Š” ์ ์—์„œ, ๋ฌธ์ œ๋ฅผ ์ˆจ๊ธฐ๊ธฐ๋ณด๋‹ค ํˆฌ๋ช…ํ•˜๊ฒŒ ๋“œ๋Ÿฌ๋‚ด๋ ค๋Š” ํƒœ๋„๊ฐ€ ๋ณด์—ฌ์š”. ๊ด€๋ จํ•ด์„œ GPT-5.6์ด ํŒŒ์ผ์„ ์˜ˆ๊ธฐ์น˜ ์•Š๊ฒŒ ์‚ญ์ œํ–ˆ๋‹ค๋Š” ๋ณด๊ณ ๋“ค์— ๋Œ€ํ•œ ์กฐ์‚ฌ ๊ฒฐ๊ณผ๋„ ๊ณต์œ ํ–ˆ๊ณ [5], ํ”„๋กฌํ”„ํŠธ ์ธ์ ์…˜(prompt injection) ์ทจ์•ฝ์ ์„ ์ž๋™์œผ๋กœ ํƒ์ง€ํ•˜๋Š” ๋‚ด๋ถ€ ๋ ˆ๋“œํŒ€ ๋„๊ตฌ ‘GPT-Red’ ๊ณต๊ฐœ๋„ ์ฃผ๋ชฉํ–ˆ์–ด์š”.[6]

AI๊ฐ€ ์ผ์ž๋ฆฌ๋ฅผ ์ˆœ(net) ์ฐฝ์ถœํ•ด์™”๋‹ค๋Š” ๊ฐœ์ธ์ ์ธ ๊ด€์ฐฐ[7]๊ณผ, ์˜์‚ฌ๋ณด๋‹ค GPT-5.6 ์‘๋‹ต์—์„œ ๊ฒฐํ•จ์ด ๋” ์ ์—ˆ๋‹ค๋Š” ์˜๋ฃŒ ์—ฐ๊ตฌ ๊ฒฐ๊ณผ ์ธ์šฉ[8]๊นŒ์ง€ โ€” ๋น ๋ฅธ ์„ฑ์žฅ ์†์—์„œ๋„ ์‚ฌํšŒ์  ์˜ํ–ฅ์„ ๊พธ์ค€ํžˆ ์˜์‹ํ•˜๋Š” ์‹œ๊ฐ์ด ์—ฟ๋ณด์—ฌ์š”.

๐Ÿ“š ์ถœ์ฒ˜ (Sources)

  1. [1]@sama on ๐• ยท 2026-07-16 โ€” “we did not have our best last 12 months ever, which is mostly my fault, but we are about tโ€ฆ”
  2. [2]@sama on ๐• ยท 2026-07-21 โ€” “we had a significant security incident during evaluation of our models. we are sharing whaโ€ฆ”
  3. [3]@sama on ๐• ยท 2026-07-14 โ€” “GPT-5.6 sol is half the price and ~twice as token efficient as fable in many cases for accโ€ฆ”
  4. [4]@sama on ๐• ยท 2026-07-14 โ€” “2.5x increase in usage of our agentic products (codex and chatgpt work) in the last week!
    โ€ฆ”
  5. [5]โ†ป RT@sama on ๐• ยท 2026-07-16 โ€” “On file deletions. Weโ€™ve investigated a handful of reports where GPT-5.6 unexpectedly deleโ€ฆ”
  6. [6]โ†ป RT@sama on ๐• ยท 2026-07-15 โ€” “Introducing GPT-Red

    An internal automated red teamer on a mission to find our modelsโ€™ proโ€ฆ”

  7. [7]@sama on ๐• ยท 2026-07-11 โ€” “so far at least, i’m pretty sure AI has been net job-creating.

    this was not what i expectโ€ฆ”

  8. [8]@sama on ๐• ยท 2026-07-11 โ€” “”physicians found fewer flaws in GPT-5.6 responses than physician-written responses.””

๐Ÿ“… 2026.06.21 ~ 07.05

AI ๋ชจ๋ธ ์ถœ์‹œ๋„ ์ •๋ถ€ ํ˜‘์˜๊ฐ€ ํ•„์š”ํ•œ ์‹œ๋Œ€๊ฐ€ ๋๋‹ค

OpenAI๊ฐ€ ์ƒˆ ๋ชจ๋ธ์„ ๊ณต๊ฐœํ•˜๋ ค ํ–ˆ์ง€๋งŒ, ๋ฏธ๊ตญ ์ •๋ถ€์˜ ์š”์ฒญ์œผ๋กœ ์ „๋ฉด ๊ณต๊ฐœ ๋Œ€์‹  ์ œํ•œ์  ํ”„๋ฆฌ๋ทฐ๋กœ ์ถœ๋ฐœํ•˜๊ฒŒ ๋์–ด์š”. ๊ฐ•๋ ฅํ•œ AI ๋ชจ๋ธ์˜ ๋ฐฐํฌ ๋ฐฉ์‹์ด ์ด์ œ ๊ธฐ์—… ๋‹จ๋…์˜ ๊ฒฐ์ •์ด ์•„๋‹Œ, ์ •๋ถ€์™€์˜ ํ˜‘์˜ ์‚ฌํ•ญ์ด ๋œ ํ˜„์‹ค์„ ๋ณด์—ฌ์ฃผ๋Š” ์‚ฌ๋ก€์ž…๋‹ˆ๋‹ค.

์ •๋ถ€ ์š”์ฒญ์œผ๋กœ ์ œํ•œ ์ถœ์‹œ๋œ ์ƒˆ ๋ชจ๋ธ, Sol๊ณผ Terra

Good new first: Sol is a smart, efficient, and a significant step forward. It is the same price as GPT-5.5. Also launching in the GPT-5.6 family is Terra, with 5.5-level performance at half the price.

Bad news: at the request of the US government, it is launching today in limited preview instead of the open access launch we were planning on. We are working with the government to get to general availability as fast as we can.

I think it is quite reasonable to roll out models–especially as they reach significant new levels of capability–in this way. It fits with our long-held strategy of iterative deployment. But this isn’t quite the process that we think is optimal.

[1]

GPT-5.6 ํŒจ๋ฐ€๋ฆฌ์— ์†ํ•˜๋Š” Sol๊ณผ Terra ๋‘ ๋ชจ๋ธ์ด ๊ณต๊ฐœ๋์–ด์š”. Sol์€ GPT-5.5์™€ ๋™์ผํ•œ ๊ฐ€๊ฒฉ, Terra๋Š” ์ ˆ๋ฐ˜ ๊ฐ€๊ฒฉ์— 5.5 ์ˆ˜์ค€์˜ ์„ฑ๋Šฅ์„ ์ œ๊ณตํ•œ๋‹ค๊ณ  ํ•ด์š”. ์ฃผ๋ชฉํ•  ์ ์€ ๋ฏธ๊ตญ ์ •๋ถ€์˜ ์š”์ฒญ์œผ๋กœ ์ „๋ฉด ๊ณต๊ฐœ ๋Œ€์‹  ์ œํ•œ ํ”„๋ฆฌ๋ทฐ๋กœ ์ถœ์‹œ๋๋‹ค๋Š” ๊ฒƒ์ด์—์š”. ์ด ๋ฐฉ์‹์ด ์™„์ „ํžˆ ์ด์ƒ์ ์ด์ง€๋Š” ์•Š์ง€๋งŒ, ๊ฐ•๋ ฅํ•œ ๋ชจ๋ธ์ผ์ˆ˜๋ก ๋‹จ๊ณ„์ ์œผ๋กœ ๋ฐฐํฌํ•˜๋Š” ์ „๋žต๊ณผ๋Š” ์ผ๋งฅ์ƒํ†ตํ•œ๋‹ค๊ณ  ์„ค๋ช…ํ–ˆ์–ด์š”.

์‚ฌ์ด๋ฒ„ ๋ณด์•ˆ๊นŒ์ง€ ํ™•์žฅ๋˜๋Š” AI ์—ญํ• 

We want to help all companies be secure, working with the USG and the security ecosystem.

*The full version of GPT-5.5-Cyber is here; state of the art performance on CyberGym.

*Patch The Planet and Codex Security will help solve security problems instead of just finding them.

[2]

GPT-5.5-Cyber๊ฐ€ ์ •์‹ ์ถœ์‹œ๋˜๋ฉฐ ์‚ฌ์ด๋ฒ„ ๋ณด์•ˆ ๋ฒค์น˜๋งˆํฌ์ธ CyberGym์—์„œ ์ตœ๊ณ  ์ˆ˜์ค€์˜ ์„ฑ๋Šฅ์„ ๊ธฐ๋กํ–ˆ์–ด์š”. ํŠนํžˆ ‘Patch The Planet’๊ณผ ‘Codex Security’๋Š” ๋ณด์•ˆ ์ทจ์•ฝ์ ์„ ์ฐพ๋Š” ๋ฐ ๊ทธ์น˜์ง€ ์•Š๊ณ  ์ง์ ‘ ํ•ด๊ฒฐ๊นŒ์ง€ ๋•๋Š”๋‹ค๋Š” ์ ์ด ์ƒˆ๋กœ์›Œ์š”. AI๊ฐ€ ๋ณด์•ˆ ๋ถ„์•ผ์—์„œ ํƒ์ง€๋ฅผ ๋„˜์–ด ๋Œ€์‘ ๋„๊ตฌ๋กœ ์ง„ํ™”ํ•˜๊ณ  ์žˆ๋‹ค๋Š” ์‹ ํ˜ธ์˜ˆ์š”.

โ†ป ๊ณต์œ ํ•œ ๊ธ€

RT @FoundationOAI: The OpenAI Foundation is joining Intercept as a founding partner through our AI resilience program.

As AI accelerates aโ€ฆ

[3]

OpenAI ์žฌ๋‹จ์ด AI ํšŒ๋ณต๋ ฅ(resilience) ํ”„๋กœ๊ทธ๋žจ์„ ํ†ตํ•ด Intercept์˜ ์ฐฝ๋ฆฝ ํŒŒํŠธ๋„ˆ๋กœ ํ•ฉ๋ฅ˜ํ•œ๋‹ค๋Š” ์†Œ์‹๋„ ๊ณต์œ ํ–ˆ์–ด์š”. AI ๊ธฐ์ˆ ์˜ ์‚ฌํšŒ์  ์˜ํ–ฅ์— ๋Œ€ํ•œ ์ฑ…์ž„์„ ์™ธ๋ถ€ ํŒŒํŠธ๋„ˆ์‹ญ์œผ๋กœ ๋„“ํ˜€๊ฐ€๋Š” ํ–‰๋ณด์˜ˆ์š”.

์ผ๋ จ์˜ ํ–‰๋ณด๋ฅผ ๋ณด๋ฉด, ๊ฐ•๋ ฅํ•œ AI ๋ชจ๋ธ์˜ ์ถœ์‹œ๋Š” ๊ธฐ์ˆ  ์™„์„ฑ๋„๋งŒํผ์ด๋‚˜ ์ •๋ถ€ยท์‚ฌํšŒ์™€์˜ ์‹ ๋ขฐ ๊ตฌ์ถ•์ด ํ•ต์‹ฌ ๋ณ€์ˆ˜๊ฐ€ ๋์Œ์„ ์•Œ ์ˆ˜ ์žˆ์–ด์š”. ๋ณด์•ˆ๊ณผ ๊ทœ์ œ, ๋‹จ๊ณ„์  ๋ฐฐํฌ๊ฐ€ ํ•˜๋‚˜์˜ ์ „๋žต์œผ๋กœ ๋ฌถ์ด๋Š” ์‹œ๋Œ€๊ฐ€ ์‹ค์ œ๋กœ ์™”๋‹ต๋‹ˆ๋‹ค.

๐Ÿ“š ์ถœ์ฒ˜ (Sources)

  1. [1]@sama on ๐• ยท 2026-06-26 โ€” “Good new first: Sol is a smart, efficient, and a significant step forward. It is the same โ€ฆ”
  2. [2]@sama on ๐• ยท 2026-06-22 โ€” “We want to help all companies be secure, working with the USG and the security ecosystem.
    โ€ฆ”
  3. [3]โ†ป RT@sama on ๐• ยท 2026-06-27 โ€” “The OpenAI Foundation is joining Intercept as a founding partner through our AI resilienceโ€ฆ”

๐Ÿ“… 2026.06.07 ~ 06.21

10๋…„์„ ๊ธฐ๋‹ค๋ฆฐ ํ˜‘์—…๊ณผ AI๊ฐ€ ํฌ๊ท€ ์งˆํ™˜ ํ™˜์ž์—๊ฒŒ ์ค„ ์ˆ˜ ์žˆ๋Š” ๊ฒƒ

Sam Altman์ด ์ตœ๊ทผ ์˜ค๋žœ ์—ผ์›์ด์—ˆ๋˜ ํ˜‘์—…์˜ ์‹œ์ž‘์„ ๊ณต๊ฐœ์ ์œผ๋กœ ์•Œ๋ ธ์–ด์š”. ๋™์‹œ์— AI ์ถ”๋ก  ๋Šฅ๋ ฅ์ด ์˜๋ฃŒ ํ˜„์žฅ์—์„œ ์–ด๋–ค ๋ณ€ํ™”๋ฅผ ๋งŒ๋“ค์–ด๋‚ผ ์ˆ˜ ์žˆ๋Š”์ง€์— ๊ด€ํ•œ ๊ธ€๋„ ์ฃผ๋ชฉํ•ด ๊ณต์œ ํ–ˆ๋‹ต๋‹ˆ๋‹ค. ์ด ๋‘ ํ๋ฆ„์€ OpenAI๊ฐ€ ์ง€๊ธˆ ์–ด๋””์— ์—๋„ˆ์ง€๋ฅผ ์ง‘์ค‘ํ•˜๊ณ  ์žˆ๋Š”์ง€๋ฅผ ์ž˜ ๋ณด์—ฌ์ค˜์š”.

10๋…„ ๋งŒ์— ์ด๋ฃจ์–ด์ง„ ํ˜‘์—…, Noam์˜ ํ•ฉ๋ฅ˜

noam is one of the people I have most wanted to work with since the very beginning of openai.

only took 10 years.

i think it will be worth the wait!

[1]

Sam Altman์€ Noam์ด OpenAI ์ฐฝ๋ฆฝ ์ดˆ๊ธฐ๋ถ€ํ„ฐ ํ•จ๊ป˜ ์ผํ•˜๊ณ  ์‹ถ์—ˆ๋˜ ์ธ๋ฌผ์ด์—ˆ๋‹ค๊ณ  ๋ฐํ˜”์–ด์š”. 10๋…„์ด๋ผ๋Š” ์‹œ๊ฐ„์ด ๊ฑธ๋ ธ์ง€๋งŒ ๊ทธ๋งŒํ•œ ๊ฐ€์น˜๊ฐ€ ์žˆ์„ ๊ฑฐ๋ผ๊ณ  ํ™•์‹ ํ•˜๋Š” ์–ด์กฐ๊ฐ€ ์ธ์ƒ์ ์ด์—์š”. ์ด์–ด์ง„ ํŠธ์œ—์—์„œ๋Š” ์œ ๋จธ๋ฅผ ์„ž์–ด ์ด๋ ‡๊ฒŒ ๋ง๋ถ™์ด๊ธฐ๋„ ํ–ˆ์–ด์š”.

We offer no explanation as to why Noams are so good at AI; we attribute their success, as all else, to divine benevolence.

[2]

“Noam์ด๋ผ๋Š” ์ด๋ฆ„์„ ๊ฐ€์ง„ ์‚ฌ๋žŒ๋“ค์ด ์™œ AI๋ฅผ ์ž˜ํ•˜๋Š”์ง€๋Š” ์„ค๋ช…ํ•˜์ง€ ์•Š๊ฒ ๋‹ค”๋Š” ๋†๋‹ด ์„ž์ธ ๋ฌธ์žฅ์œผ๋กœ, ์ƒˆ ๋™๋ฃŒ์— ๋Œ€ํ•œ ๊ธฐ๋Œ€์™€ ๋ฐ˜๊ฐ€์›€์„ ์œ ์พŒํ•˜๊ฒŒ ํ‘œํ˜„ํ–ˆ์–ด์š”. ๊ณต๊ฐœ ๋ฐœํ‘œ์ž„์—๋„ ๊ฐœ์ธ์ ์ธ ์˜จ๊ธฐ๊ฐ€ ๋А๊ปด์ง€๋Š” ๋ฉ”์‹œ์ง€์˜ˆ์š”.

AI ์ถ”๋ก ์ด ํฌ๊ท€ ์งˆํ™˜ ํ™˜์ž์—๊ฒŒ ๊ฐ€์žฅ ๋น›๋‚  ์ˆ˜ ์žˆ๋‹ค

โ†ป ๊ณต์œ ํ•œ ๊ธ€

RT @thekaransinghal: One of the ways scaling test-time compute can benefit people most: have reasoning models think really hard about rareโ€ฆ

[3]

Sam Altman์ด ์ฃผ๋ชฉํ•ด ๊ณต์œ ํ•œ ์ด ๊ธ€์€, ํ…Œ์ŠคํŠธ ํƒ€์ž„ ์ปดํ“จํŠธ(test-time compute, ๋ชจ๋ธ์ด ๋‹ต์„ ๋‚ด๊ธฐ ์ „์— ๋” ๋งŽ์€ ์—ฐ์‚ฐ์„ ํˆฌ์ž…ํ•˜๋Š” ๋ฐฉ์‹) ํ™•์žฅ์˜ ๊ฐ€์žฅ ํฐ ์ˆ˜ํ˜œ์ž๊ฐ€ ํฌ๊ท€ ์งˆํ™˜ ํ™˜์ž๊ฐ€ ๋  ์ˆ˜ ์žˆ๋‹ค๋Š” ๊ด€์ ์„ ๋‹ด๊ณ  ์žˆ์–ด์š”. ์ง„๋‹จ ์‚ฌ๋ก€๊ฐ€ ์ ๊ณ  ์ „๋ฌธ์˜๋„ ๋ถ€์กฑํ•œ ํฌ๊ท€ ์งˆํ™˜์ผ์ˆ˜๋ก, ์ถ”๋ก  ๋ชจ๋ธ์ด ๊นŠ์ด ์ƒ๊ฐํ•  ์‹œ๊ฐ„์„ ์ฃผ๋ฉด ์‹ค์งˆ์ ์ธ ๋„์›€์ด ๋  ์ˆ˜ ์žˆ๋‹ค๋Š” ๊ฑฐ์˜ˆ์š”. ์˜๋ฃŒ AI์˜ ๊ฐ€์น˜๊ฐ€ ๊ฐ€์žฅ ํ•„์š”ํ•œ ๊ณณ์—์„œ ๊ฐ€์žฅ ํฌ๊ฒŒ ๋ฐœํœ˜๋  ์ˆ˜ ์žˆ๋‹ค๋Š” ์‹œ๊ฐ์ด์—์š”.

์ด๋ฒˆ ํ™œ๋™์—์„œ Sam Altman์€ ์˜ค๋žœ ์ธ์—ฐ์˜ ๊ฒฐ์‹ค๊ณผ AI๊ฐ€ ์‚ฌํšŒ์ ์œผ๋กœ ๊ธฐ์—ฌํ•  ์ˆ˜ ์žˆ๋Š” ๊ตฌ์ฒด์ ์ธ ์ง€์ ์„ ํ•จ๊ป˜ ์งš์—ˆ์–ด์š”. ๊ธฐ์ˆ ์˜ ์ง„๋ณด๋ฅผ ์‚ฌ๋žŒ๊ณผ ํ˜„์‹ค ๋ฌธ์ œ์— ์—ฐ๊ฒฐํ•˜๋ ค๋Š” ์‹œ์„ ์ด ์ผ๊ด€๋˜๊ฒŒ ๋А๊ปด์ ธ์š”.

๐Ÿ“š ์ถœ์ฒ˜ (Sources)

  1. [1]@sama on ๐• ยท 2026-06-18 โ€” “noam is one of the people I have most wanted to work with since the very beginning of openโ€ฆ”
  2. [2]@sama on ๐• ยท 2026-06-18 โ€” “We offer no explanation as to why Noams are so good at AI; we attribute their success, as โ€ฆ”
  3. [3]โ†ป RT@sama on ๐• ยท 2026-06-18 โ€” “One of the ways scaling test-time compute can benefit people most: have reasoning models tโ€ฆ”

๐Ÿ“… 2026.05.24 ~ 06.07

AI๋Š” ๋””์ง€ํ„ธ์„ ๋„˜์–ด ๋ฌผ๋ฆฌ ์„ธ๊ณ„๋กœ ํ™•์žฅ๋˜๊ณ , ์‚ฌํšŒ ์ „์ฒด์˜ ํšŒ๋ณต๋ ฅ์„ ํ‚ค์›Œ์•ผ ํ•œ๋‹ค

Sam Altman์€ ์ตœ๊ทผ AI์˜ ์—ญํ• ์„ ๋‹จ์ˆœํ•œ ์†Œํ”„ํŠธ์›จ์–ด ๋„๊ตฌ์—์„œ ๋ฌผ๋ฆฌ ์„ธ๊ณ„์˜ ์ธํ”„๋ผ๋กœ ํ™•์žฅํ•˜๋Š” ๊ตฌ์ƒ์„ ์ž‡๋”ฐ๋ผ ๊ณต๊ฐœํ–ˆ์–ด์š”. ๋™์‹œ์— AI๊ฐ€ ๊ฐ€์ ธ์˜ฌ ์‚ฌํšŒ์  ์ถฉ๊ฒฉ์— ๋Œ€๋น„ํ•˜๊ธฐ ์œ„ํ•œ ์žฌ๋‹จ ์„ค๋ฆฝ, ์ƒ๋ฌผ๋ฐฉ์–ด ์ด๋‹ˆ์…”ํ‹ฐ๋ธŒ, ์ •๋ถ€ ํ–‰์ •๋ช…๋ น ์ง€์ง€ ๋“ฑ ๊ฑฐ๋ฒ„๋„Œ์Šค ์ธก๋ฉด์—์„œ๋„ ์ ๊ทน์ ์ธ ํ–‰๋ณด๋ฅผ ๋ณด์ด๊ณ  ์žˆ์–ด์š”.

๋กœ๋ณดํ‹ฑ์Šค: AI๊ฐ€ ๋ฌผ๋ฆฌ ์„ธ๊ณ„๋กœ ๋“ค์–ด์˜จ๋‹ค

OpenAI Robotics is hiring, looking for exceptional full-stack hardware, ops, systems, and ML engineers to help us program and manufacture robots that are useful for society.

AI should be able to help people in the physical world. In the short term, we are focused on robots to support skilled workers to build our future infrastructure; in the long term, we imagine everyone having a personal robot doing anything they need.

[1]

OpenAI๊ฐ€ ๊ณต์‹์ ์œผ๋กœ ๋กœ๋ณดํ‹ฑ์Šค ํŒ€์„ ์ถœ๋ฒ”์‹œ์ผฐ์–ด์š”. ์„ธ๊ณ„ ์‹œ๋ฎฌ๋ ˆ์ด์…˜(world simulation) ์—ฐ๊ตฌ ํ”„๋กœ๊ทธ๋žจ์ด 1๋…„์—ฌ ๋งŒ์— OpenAI Robotics๋กœ ๋ฐœ์ „ํ–ˆ๊ณ , ํ•˜๋“œ์›จ์–ด์™€ ML(๋จธ์‹ ๋Ÿฌ๋‹) ์—ฐ๊ตฌ์˜ ๊ณต๋™ ์„ค๊ณ„๋ฅผ ๊ธฐ๋ฐ˜์œผ๋กœ ๋น ๋ฅด๊ฒŒ ์ง„์ฒ™ ์ค‘์ด๋ผ๊ณ  ํ•ด์š”. ๋‹จ๊ธฐ์ ์œผ๋กœ๋Š” ์ˆ™๋ จ ๋…ธ๋™์ž๋ฅผ ๋ณด์กฐํ•˜๋Š” ๋กœ๋ด‡, ์žฅ๊ธฐ์ ์œผ๋กœ๋Š” ๋ชจ๋“  ์‚ฌ๋žŒ์ด ๊ฐœ์ธ ๋กœ๋ด‡์„ ๊ฐ–๋Š” ๋ฏธ๋ž˜๋ฅผ ๋ชฉํ‘œ๋กœ ์ œ์‹œํ–ˆ์–ด์š”.

โ†ป ๊ณต์œ ํ•œ ๊ธ€

RT @thsottiaux: Five million users would agree. Resetting the limits tomorrow morning to celebrate.

Time to go /fast

[2]

๊ด€๋ จํ•ด 500๋งŒ ์‚ฌ์šฉ์ž ๋ŒํŒŒ๋ฅผ ์ž์ถ•ํ•˜๋Š” ๊ธ€๋„ ๊ณต์œ ํ–ˆ์–ด์š”. AI ์ œํ’ˆ์ด ์‹ค์ œ ์‚ฌ์šฉ์ž ๊ทœ๋ชจ๋กœ ๋น ๋ฅด๊ฒŒ ํ™•์‚ฐ๋˜๊ณ  ์žˆ๋Š” ํ๋ฆ„์„ ํ•จ๊ป˜ ์ฃผ๋ชฉํ•œ ์…ˆ์ด์—์š”.

AI์˜ ์‚ฌํšŒ์  ์ฑ…์ž„: ๋ฒˆ์˜, ๋ฐฉ์–ด, ๊ทธ๋ฆฌ๊ณ  ํšŒ๋ณต๋ ฅ

AI should dramatically increase quality of life and individual freedoms for people around the world.

The OpenAI Foundation is making an initial $250M commitment to measurement, transition support, and new approaches to broadly shared prosperity.

[3]

OpenAI ์žฌ๋‹จ์ด 2์–ต 5์ฒœ๋งŒ ๋‹ฌ๋Ÿฌ(์•ฝ 3,400์–ต ์›) ๊ทœ๋ชจ์˜ ์ดˆ๊ธฐ ํˆฌ์ž๋ฅผ ๋ฐœํ‘œํ–ˆ์–ด์š”. ์‚ถ์˜ ์งˆ ์ธก์ •, ์ „ํ™˜๊ธฐ ์ง€์›, ๊ด‘๋ฒ”์œ„ํ•œ ๊ณต๋™ ๋ฒˆ์˜ ๋ชจ๋ธ ์—ฐ๊ตฌ๊ฐ€ ํ•ต์‹ฌ ๊ณผ์ œ์˜ˆ์š”. [4] ์ƒ๋ฌผ๋ฐฉ์–ด(biodefense) ๋ถ„์•ผ์—์„œ๋„ ์„ธ๊ณ„๊ฐ€ ๋ฏธ๋ฆฌ ์ค€๋น„ํ•  ์ˆ˜ ์žˆ๋„๋ก ๋•๊ฒ ๋‹ค๋Š” ์˜์ง€๋ฅผ ๋ณ„๋„๋กœ ๋ฐํ˜”๊ณ [5] ๋ฏธ๊ตญ์˜ AI ์‚ฌ์ด๋ฒ„ ๋„๊ตฌ๋ฅผ ์‹ ๋ขฐํ•  ์ˆ˜ ์žˆ๋Š” ๋ฐฉ์–ด์ž๋“ค์—๊ฒŒ ์ œ๊ณตํ•˜๋Š” ๋ฐฉํ–ฅ์˜ ํ–‰์ •๋ช…๋ น(EO)์—๋„ ์ง€์ง€ ์ž…์žฅ์„ ํ‘œ๋ช…ํ–ˆ์–ด์š”.[6]

Sam Altman์˜ ์ตœ๊ทผ ํ–‰๋ณด๋Š” AI๋ฅผ “๋” ์ข‹์€ ์•ฑ”์ด ์•„๋‹Œ ์‚ฌํšŒ ์ธํ”„๋ผ๋กœ ๋ณด๋Š” ์‹œ๊ฐ์„ ๋šœ๋ ท์ด ๋ณด์—ฌ์ค˜์š”. ๊ธฐ์ˆ  ๊ฐœ๋ฐœ๊ณผ ์‚ฌํšŒ์  ๋Œ€๋น„๋ฅผ ๋™์‹œ์— ์ง„ํ–‰ํ•ด์•ผ ํ•œ๋‹ค๋Š” ๋ฉ”์‹œ์ง€๊ฐ€ ์ผ๊ด€๋˜๊ฒŒ ๋‹ด๊ฒจ ์žˆ์–ด์š”.

๐Ÿ“š ์ถœ์ฒ˜ (Sources)

  1. [1]@sama on ๐• ยท 2026-05-31 โ€” “OpenAI Robotics is hiring, looking for exceptional full-stack hardware, ops, systems, and โ€ฆ”
  2. [2]โ†ป RT@sama on ๐• ยท 2026-05-31 โ€” “Five million users would agree. Resetting the limits tomorrow morning to celebrate.

    Time โ€ฆ”

  3. [3]@sama on ๐• ยท 2026-05-27 โ€” “AI should dramatically increase quality of life and individual freedoms for people around โ€ฆ”
  4. [4]@sama on ๐• ยท 2026-06-01 โ€” “The OpenAI Foundation is doing a lot of wonderful things.

    Helping society become resilienโ€ฆ”

  5. [5]@sama on ๐• ยท 2026-05-31 โ€” “We want to help the world get a head start on biodefense:”
  6. [6]@sama on ๐• ยท 2026-06-03 โ€” “theUSshould lead on AI by continuing to develop the very best models, making sure they’re โ€ฆ”

๐Ÿ“… 2026.05.10 ~ 05.24

AGI๊ฐ€ ์ˆ˜ํ•™ ๋‚œ์ œ๋ฅผ ํ’€๊ณ , ์ฝ”๋“œ๋ฅผ ์งœ๊ณ , ์‚ฌ์ด๋ฒ„๋ฅผ ๋ฐฉ์–ดํ•œ๋‹ค โ€” OpenAI๊ฐ€ ๊ทธ๋ฆฌ๋Š” ์„ธ ๊ฐ€์ง€ ์ „์„ 

๋ฒ”์šฉ AI ๋ชจ๋ธ์ด ์ˆ˜์‹ญ ๋…„ ๋ฌต์€ ์ˆ˜ํ•™ ๋‚œ์ œ๋ฅผ ํ•ด๊ฒฐํ•˜๊ณ , ์ฝ”๋”ฉ ์—์ด์ „ํŠธ๊ฐ€ ๋‚ฎ์ž  ์‹œ๊ฐ„ ๋™์•ˆ ํƒœ์Šคํฌ๋ฅผ ์ฒ˜๋ฆฌํ•˜๋ฉฐ, AI๊ฐ€ ๊ธฐ์—… ๋ณด์•ˆ์„ ์‹ค์‹œ๊ฐ„์œผ๋กœ ์ง€์ผœ๋‚ด๋Š” ์‹œ๋Œ€๊ฐ€ ๋„๋ž˜ํ–ˆ์–ด์š”. Sam Altman์˜ ์ตœ๊ทผ ํ”ผ๋“œ๋Š” ์ด ์„ธ ํ๋ฆ„์ด ๋™์‹œ์— ๊ฐ€์†๋˜๊ณ  ์žˆ์Œ์„ ๋ณด์—ฌ์ค˜์š”.

AGI, ์—ฐ๊ตฌ ์ตœ์ „์„ ์„ ๋ฐ”๊พธ๋‹ค

a general-purpose model solved a major open problem in mathematics.

we’ll be saying this a lot over the coming years, but this is a kinda big milestone.

i’m very excited for AI to greatly extend our understanding of the world, but still, i have complicated feelings today.

[1]

๋ฒ”์šฉ ๋ชจ๋ธ์ด ์กฐํ•ฉ๋ก (combinatorics) ๋ถ„์•ผ์˜ ์˜ค๋žœ ๋ฏธํ•ด๊ฒฐ ๋ฌธ์ œ์ธ ‘๋‹จ์œ„ ๊ฑฐ๋ฆฌ(unit distance)’ ๋ฌธ์ œ๋ฅผ ๋ŒํŒŒํ–ˆ์–ด์š”. Altman์€ ๊ธฐ์จ๊ณผ ๋™์‹œ์— “๋ณต์žกํ•œ ๊ฐ์ •”์„ ๋А๋‚€๋‹ค๊ณ  ์†”์งํ•˜๊ฒŒ ํ„ธ์–ด๋†“์•˜๋Š”๋ฐ, ์ด๋Š” AI๊ฐ€ ์ธ๊ฐ„ ์ง€์‹์˜ ๊ฒฝ๊ณ„๋ฅผ ์žฌํŽธํ•˜๋Š” ์ˆœ๊ฐ„์ด ์ฃผ๋Š” ๋ฌด๊ฒŒ๊ฐ์„ ๋“œ๋Ÿฌ๋‚ด์š”. ๊ทธ๋Š” AGI๊ฐ€ ์—ฐ๊ตฌ, ๊ธฐ์—…, ๊ทธ๋ฆฌ๊ณ  ๊ฐœ์ธ์˜ ๋ชฉํ‘œ ๋‹ฌ์„ฑ์„ ๊ฐ€์†ํ•˜๋Š” ์„ธ ๋ฐฉํ–ฅ์„ ํ•ต์‹ฌ ๊ณผ์ œ๋กœ ๊ผฝ๊ธฐ๋„ ํ–ˆ์–ด์š”.[5] ๊ด€๋ จํ•ด, ํ•ด๋‹น ๋ชจ๋ธ์„ ํ›ˆ๋ จํ•œ ์—ฐ๊ตฌ์›์˜ ๊ธ€๋„ ๊ณต์œ ํ•˜๋ฉฐ ์ด ์ˆœ๊ฐ„์˜ ์˜๋ฏธ๋ฅผ ํ•จ๊ป˜ ๊ฐ•์กฐํ–ˆ์–ด์š”.

โ†ป ๊ณต์œ ํ•œ ๊ธ€

RT @SherylHsu02: It’s a really special time to be aliveโ€ฆsome thoughts from training this model ๐Ÿงต

[2]

๋ชจ๋ธ์„ ์ง์ ‘ ํ›ˆ๋ จํ•œ ์—ฐ๊ตฌ์›์˜ ๊ฐํšŒ๋ฅผ ๊ณต์œ ํ•˜๋ฉฐ, Altman์€ ์ด ๋ŒํŒŒ๊ตฌ๊ฐ€ ํŒ€ ์ „์ฒด์˜ ์„ฑ๊ณผ์ž„์„ ๊ฐ„์ ‘์ ์œผ๋กœ ํ‘œํ˜„ํ–ˆ์–ด์š”.

Codex๋กœ ๊ฐœ๋ฐœ์ž๋ฅผ ‘์Šˆํผํžˆ์–ด๋กœ’๋กœ, ๋ณด์•ˆ์€ Daybreak๋กœ

way cooler to help software developers pokemon-evolve into superheroes than to try to replace them

it is insane what one really good person can do now

[3]

Altman์€ AI๊ฐ€ ๊ฐœ๋ฐœ์ž๋ฅผ ๋Œ€์ฒดํ•˜๋Š” ๊ฒƒ์ด ์•„๋‹ˆ๋ผ ๋Šฅ๋ ฅ์„ ๊ทน๋Œ€ํ™”ํ•˜๋Š” ๋„๊ตฌ์—ฌ์•ผ ํ•œ๋‹ค๊ณ  ๊ฐ•์กฐํ•ด์š”. Codex๋ฅผ ์‹ค์ œ๋กœ ์จ๋ณด๋ฉฐ “ํƒœ์Šคํฌ๋ฅผ ๊ฑธ์–ด๋†“๊ณ  ์•„์ด์™€ ๋†€๋‹ค ๋Œ์•„์˜ค๋‹ˆ ์™„๋ฃŒ๋ผ ์žˆ์—ˆ๋‹ค”๊ณ  ์ „ํ•˜๊ธฐ๋„ ํ–ˆ๋Š”๋ฐ,[6] ์ด๋Š” AI ์—์ด์ „ํŠธ๊ฐ€ ์ด๋ฏธ ์ผ์ƒ์  ์›Œํฌํ”Œ๋กœ์šฐ์— ๋…น์•„๋“ค์—ˆ์Œ์„ ๋ณด์—ฌ์ค˜์š”. ํ•œํŽธ OpenAI๋Š” ์‚ฌ์ด๋ฒ„ ๋ณด์•ˆ ์ด๋‹ˆ์…”ํ‹ฐ๋ธŒ ‘Daybreak’๋ฅผ ๋ก ์นญํ•ด ๊ธฐ์—…๋“ค์ด AI ๊ธฐ๋ฐ˜์œผ๋กœ ์†Œํ”„ํŠธ์›จ์–ด๋ฅผ ์ง€์† ๋ณดํ˜ธ๋ฐ›์„ ์ˆ˜ ์žˆ๋„๋ก ์ง€์›ํ•˜๊ฒ ๋‹ค๊ณ  ๋ฐํ˜”์–ด์š”.[7] ์•„์šธ๋Ÿฌ AI ์—์ด์ „ํŠธ์˜ ์˜ค์ •๋ ฌ(misalignment)์„ ๋ง‰๊ธฐ ์œ„ํ•œ ‘์‚ฌ๊ณ  ์—ฐ์‡„ ๋ชจ๋‹ˆํ„ฐ(chain of thought monitor)’ ๊ธฐ์ˆ ๋„ ์ฃผ๋ชฉํ•  ๋งŒํ•ด์š”.

โ†ป ๊ณต์œ ํ•œ ๊ธ€

RT @OpenAI: Chain of thought monitors are a key layer of defense against AI agent misalignment. To preserve monitorability, we avoid penalizing the model for its reasoning.

[4]

AI ์—์ด์ „ํŠธ๊ฐ€ ๊ฐ•๋ ฅํ•ด์งˆ์ˆ˜๋ก ์•ˆ์ „์žฅ์น˜๋„ ํ•จ๊ป˜ ์ง„ํ™”ํ•ด์•ผ ํ•œ๋‹ค๋Š” ๋ฉ”์‹œ์ง€์˜ˆ์š”. Altman์˜ ํ”ผ๋“œ ์ „๋ฐ˜์€ “๋” ๊ฐ•๋ ฅํ•œ AI”์™€ “๋” ์•ˆ์ „ํ•œ AI”๋ฅผ ๋™์‹œ์— ์ถ”๊ตฌํ•˜๋Š” ๊ธด์žฅ๊ฐ์„ ์ผ๊ด€๋˜๊ฒŒ ๋‹ด๊ณ  ์žˆ์–ด์š”.

์ˆ˜ํ•™ ๋‚œ์ œ ํ•ด๊ฒฐ๋ถ€ํ„ฐ ์ฝ”๋”ฉ ์—์ด์ „ํŠธ, ์‚ฌ์ด๋ฒ„ ๋ฐฉ์–ด๊นŒ์ง€ โ€” Altman์ด ๊ทธ๋ฆฌ๋Š” AGI ์‹ค์šฉํ™”๋Š” ํŠน์ • ๋ถ„์•ผ๊ฐ€ ์•„๋‹Œ ์‚ฌํšŒ ์ „๋ฐ˜์— ๊ฑธ์ณ ๋™์‹œ๋‹ค๋ฐœ์ ์œผ๋กœ ์ง„ํ–‰๋˜๊ณ  ์žˆ์–ด์š”. ๋Šฅ๋ ฅ์˜ ํ™•์žฅ๊ณผ ์•ˆ์ „์˜ ํ™•๋ณด, ๋‘ ์ถ•์„ ํ•จ๊ป˜ ์žก์œผ๋ ค๋Š” ๊ณ ๋ฏผ์ด ๊ทธ์˜ ๊ธ€ ๊ณณ๊ณณ์— ๋ฐฐ์–ด ์žˆ๋‹ต๋‹ˆ๋‹ค.

๐Ÿ“š ์ถœ์ฒ˜ (Sources)

  1. [1]@sama on ๐• ยท 2026-05-20 โ€” “a general-purpose model solved a major open problem in mathematics.

    we’ll be saying this โ€ฆ”

  2. [2]โ†ป RT@sama on ๐• ยท 2026-05-20 โ€” “Itโ€™s a really special time to be aliveโ€ฆsome thoughts from training this model ๐Ÿงต”
  3. [3]@sama on ๐• ยท 2026-05-07 โ€” “way cooler to help software developers pokemon-evolve into superheroes than to try to replโ€ฆ”
  4. [4]โ†ป RT@sama on ๐• ยท 2026-05-08 โ€” “Chain of thought monitors are a key layer of defense against AI agent misalignment. To preโ€ฆ”
  5. [5]@sama on ๐• ยท 2026-05-20 โ€” “three of the things we are most excited about:

    1. AGI accelerating research
    2. AGI acceleโ€ฆ”

  6. [6]@sama on ๐• ยท 2026-05-09 โ€” “kicking off a bunch of codex tasks, running around with my kid in the sunshine, and then cโ€ฆ”
  7. [7]@sama on ๐• ยท 2026-05-11 โ€” “OpenAI is launching Daybreak, our effort to accelerate cyber defense and continuously secuโ€ฆ”

๐Ÿ“… 2026.04.06 ~ 04.20

AI ์ฝ”๋”ฉ ๋„๊ตฌ๊ฐ€ ํ„ฐ๋ฏธ๋„์„ ๋ฒ—์–ด๋‚˜ Mac ์ „์ฒด๋ฅผ ์›€์ง์ด๊ธฐ ์‹œ์ž‘ํ–ˆ๋‹ค

OpenAI์˜ ์ฝ”๋”ฉ ์—์ด์ „ํŠธ Codex๊ฐ€ ๋น ๋ฅด๊ฒŒ ์‚ฌ์šฉ์ž๋ฅผ ๋Œ์–ด๋ชจ์œผ๊ณ  ์žˆ์–ด์š”. ์ฃผ๊ฐ„ ํ™œ์„ฑ ์‚ฌ์šฉ์ž 300๋งŒ ๋ช…์„ ๋ŒํŒŒํ–ˆ๊ณ , ๋‹จ์ˆœํ•œ ์ฝ”๋“œ ์ž‘์„ฑ ๋„๊ตฌ๋ฅผ ๋„˜์–ด Mac ์•ฑ ์ „์ฒด๋ฅผ ์ง์ ‘ ์กฐ์ž‘ํ•˜๋Š” ์ˆ˜์ค€์œผ๋กœ ์ง„ํ™”ํ–ˆ์Šต๋‹ˆ๋‹ค. ๊ฐ™์€ ์‹œ๊ธฐ OpenAI๋Š” ์ƒ๋ฌผ์˜ํ•™ ์—ฐ๊ตฌ ์ „์šฉ ์ถ”๋ก  ๋ชจ๋ธ๊นŒ์ง€ ๊ณต๊ฐœํ•˜๋ฉฐ AI ์ ์šฉ ๋ฒ”์œ„๋ฅผ ๋น ๋ฅด๊ฒŒ ๋„“ํžˆ๊ณ  ์žˆ์–ด์š”.

ํ„ฐ๋ฏธ๋„ ๋ฐ–์œผ๋กœ ๋‚˜์˜จ Codex โ€” ์ปดํ“จํ„ฐ ์ „์ฒด๊ฐ€ ์ž‘์—… ๊ณต๊ฐ„์ด ๋˜๋‹ค

โ†ป ๊ณต์œ ํ•œ ๊ธ€

This is the first time I’ve ever seen an LLM operate a GUI as fast as a person, and it’s surreal.

[1]

Sam Altman์ด ์ด ๊ธ€์— ์ฃผ๋ชฉํ–ˆ์–ด์š”. LLM(๊ฑฐ๋Œ€ ์–ธ์–ด ๋ชจ๋ธ)์ด ๊ทธ๋ž˜ํ”ฝ ์ธํ„ฐํŽ˜์ด์Šค(GUI)๋ฅผ ์‚ฌ๋žŒ๋งŒํผ ๋น ๋ฅด๊ฒŒ ๋‹ค๋ฃจ๋Š” ์žฅ๋ฉด์„ ์ฒ˜์Œ ๋ชฉ๊ฒฉํ–ˆ๋‹ค๋Š” ๋ฐ˜์‘์ด์—์š”. Codex์˜ ‘์ปดํ“จํ„ฐ ์‚ฌ์šฉ(Computer Use)’ ๊ธฐ๋Šฅ์ด ์‹ค์ œ๋กœ ๊ตฌํ˜„๋œ ์ˆœ๊ฐ„์„ ๋‹ด์€ ์˜์ƒ์ž…๋‹ˆ๋‹ค.

Computer use is a real update for me; it feels even more useful than I expected. It can use all of the apps on your Mac, in parallel and without interfering with your direct work.

[2]

Codex์˜ ์ปดํ“จํ„ฐ ์‚ฌ์šฉ ๊ธฐ๋Šฅ์ด Mac์˜ ๋ชจ๋“  ์•ฑ์„ ๋ณ‘๋ ฌ๋กœ ์‹คํ–‰ํ•˜๋ฉด์„œ๋„ ์‚ฌ์šฉ์ž์˜ ์ž‘์—…์„ ๋ฐฉํ•ดํ•˜์ง€ ์•Š๋Š”๋‹ค๊ณ  ์ง์ ‘ ํ‰๊ฐ€ํ–ˆ์–ด์š”. ‘์˜ˆ์ƒ๋ณด๋‹ค ํ›จ์”ฌ ์œ ์šฉํ•˜๋‹ค’๋Š” ํ‘œํ˜„์—์„œ ๋‚ด๋ถ€์ ์œผ๋กœ๋„ ๊ธฐ๋Œ€์น˜๋ฅผ ๋›ฐ์–ด๋„˜์€ ์—…๋ฐ์ดํŠธ์ž„์„ ์•Œ ์ˆ˜ ์žˆ์–ด์š”. ๊ฐ™์€ ๋‚  ์ธ์•ฑ ๋ธŒ๋ผ์šฐ์ €์™€ ์ฝ”๋ฉ˜ํŠธ ๋ชจ๋“œ ์ถœ์‹œ ์†Œ์‹๋„ ํ•จ๊ป˜ ๊ณต์œ ํ–ˆ์–ด์š”.[6]

โ†ป ๊ณต์œ ํ•œ ๊ธ€

Codex โ€” Compute efficient โœ… Always up, never down โœ… Best at hardcore engineering โœ… Crazy good app, first to escape the terminalโ€ฆ

[3]

์‚ฌ์šฉ์ž๋“ค์ด Codex๋ฅผ ‘ํ„ฐ๋ฏธ๋„์„ ์ฒ˜์Œ์œผ๋กœ ํƒˆ์ถœํ•œ AI’๋กœ ํ‰๊ฐ€ํ•œ ๊ธ€๋„ ๊ณต์œ ํ–ˆ์–ด์š”. ๊ณ ์„ฑ๋Šฅ ์—”์ง€๋‹ˆ์–ด๋ง ์ž‘์—…์—์„œ ์•ˆ์ •์„ฑ๊ณผ ํšจ์œจ์„ ๋™์‹œ์— ๊ฐ–์ท„๋‹ค๋Š” ํ˜„์žฅ ๋ฐ˜์‘์ž…๋‹ˆ๋‹ค. 300๋งŒ ๋ช… ๋ŒํŒŒ๋ฅผ ๊ธฐ๋…ํ•ด ์‚ฌ์šฉ๋Ÿ‰ ์ œํ•œ์„ ์ดˆ๊ธฐํ™”ํ•˜๊ณ , ์ดํ›„ 100๋งŒ ๋ช… ๋‹จ์œ„๋กœ ๋ฐ˜๋ณตํ•˜๊ฒ ๋‹ค๊ณ  ๋ฐํ˜”์–ด์š”.[7]

์ƒ๋ฌผ์˜ํ•™๋ถ€ํ„ฐ ์•Œ์ธ ํ•˜์ด๋จธ๊นŒ์ง€ โ€” AI๊ฐ€ ๊ณผํ•™ ์—ฐ๊ตฌ์— ์ง„์ž…ํ•˜๋‹ค

โ†ป ๊ณต์œ ํ•œ ๊ธ€

RT @OpenAI: Introducing GPT-Rosalind, our frontier reasoning model built to support research across biology, drug discovery, and translationโ€ฆ

[4]

Sam Altman์ด OpenAI์˜ ๊ณต์‹ ๋ฐœํ‘œ๋ฅผ ๊ณต์œ ํ–ˆ์–ด์š”. GPT-Rosalind๋Š” ์ƒ๋ฌผํ•™, ์‹ ์•ฝ ๊ฐœ๋ฐœ, ๋ฒˆ์—ญ ์—ฐ๊ตฌ๋ฅผ ์ง€์›ํ•˜๋Š” ์ตœ์ „์„  ์ถ”๋ก  ๋ชจ๋ธ(frontier reasoning model)๋กœ, DNA ์ด์ค‘๋‚˜์„ ์„ ๋ฐœ๊ฒฌํ•œ ๋กœ์ž˜๋ฆฐ๋“œ ํ”„๋žญํด๋ฆฐ์˜ ์ด๋ฆ„์—์„œ ๋”ฐ์™”์–ด์š”.

โ†ป ๊ณต์œ ํ•œ ๊ธ€

Alzheimer’s is one of medicine’s hardest unsolved problems, and one of the most devastating. At the OpenAI Foundation, we bโ€ฆ

[5]

OpenAI ์žฌ๋‹จ์ด ์•Œ์ธ ํ•˜์ด๋จธ ์—ฐ๊ตฌ์— AI๋ฅผ ์ ‘๋ชฉํ•˜๋Š” ํ”„๋กœ์ ํŠธ๋ฅผ ์ถ”์ง„ ์ค‘์ด๋ผ๋Š” ๊ธ€๋„ ์ฃผ๋ชฉํ–ˆ์–ด์š”. ์ฝ”๋”ฉ ๋„๊ตฌ์™€ ์˜ํ•™ ์—ฐ๊ตฌ๋ผ๋Š” ๋‘ ์ถ•์ด ๋™์‹œ์— ์›€์ง์ด๊ณ  ์žˆ์Œ์„ ๋ณด์—ฌ์ฃผ๋Š” ํ๋ฆ„์ด์—์š”.

์ด๋ฒˆ ํ–‰๋ณด๋Š” Codex๋ฅผ ๋‹จ์ˆœ ๊ฐœ๋ฐœ์ž ๋„๊ตฌ์—์„œ ๋ฒ”์šฉ ์—์ด์ „ํŠธ๋กœ ์ „ํ™˜ํ•˜๋Š” ๋ฐฉํ–ฅ๊ณผ, AI๋ฅผ ๊ณผํ•™ ๋‚œ์ œ ํ•ด๊ฒฐ์˜ ์‹ค์งˆ์  ์ˆ˜๋‹จ์œผ๋กœ ์‚ผ์œผ๋ ค๋Š” ์˜๋„๊ฐ€ ํ•จ๊ป˜ ์ง„ํ–‰๋˜๊ณ  ์žˆ์Œ์„ ๋ณด์—ฌ์ค˜์š”. ๋‘ ๋ฐฉํ–ฅ ๋ชจ๋‘ ‘์‚ฌ์šฉ์ž ์ˆ˜’์™€ ‘์‹ค์‚ฌ์šฉ ํ”ผ๋“œ๋ฐฑ’์„ ํ•ต์‹ฌ ๋™๋ ฅ์œผ๋กœ ์‚ผ๊ณ  ์žˆ๋‹ค๋Š” ์ ์ด ์ธ์ƒ์ ์ž…๋‹ˆ๋‹ค.

๐Ÿ“š ์ถœ์ฒ˜ (Sources)

  1. [1]โ†ป RT@sama on ๐• ยท 2026-04-16 โ€” “This is the first time I’ve ever seen an LLM operate a GUI as fast as a person, and it’s sโ€ฆ”
  2. [2]@sama on ๐• ยท 2026-04-16 โ€” “Lots of major improvements to Codex!

    Computer use is a real update for me; it feels even โ€ฆ”

  3. [3]โ†ป RT@sama on ๐• ยท 2026-04-16 โ€” “Codex

    Compute efficient โœ…
    Always up, never down โœ…
    Best at hardcore engineering โœ…
    Crazy goโ€ฆ”

  4. [4]โ†ป RT@sama on ๐• ยท 2026-04-16 โ€” “Introducing GPT-Rosalind, our frontier reasoning model built to support research across biโ€ฆ”
  5. [5]โ†ป RT@sama on ๐• ยท 2026-04-08 โ€” “Alzheimerโ€™s is one of medicine’s hardest unsolved problems, and one of the most devastatinโ€ฆ”
  6. [6]โ†ป RT@sama on ๐• ยท 2026-04-16 โ€” “We are super excited to launch the in-app browser inside Codex with comment mode!

    Viewโ€ฆ”

  7. [7]@sama on ๐• ยท 2026-04-07 โ€” “To celebrate 3 million weekly codex users, we are resetting usage limits.

    We will do thisโ€ฆ”


๐Ÿ“… 2026.03.23 ~ 04.06

AI๊ฐ€ ์‹คํ—˜์‹ค ๋ฐ–์œผ๋กœ ๋‚˜์˜ค๋Š” ์‹œ๋Œ€, OpenAI๊ฐ€ ์ค€๋น„ํ•˜๋Š” ๋‘ ๊ฐ€์ง€ ํฐ ๊ทธ๋ฆผ

OpenAI๊ฐ€ ๋‹จ์ˆœํ•œ AI ๋ชจ๋ธ ํšŒ์‚ฌ๋ฅผ ๋„˜์–ด ์‚ฌํšŒ ์ธํ”„๋ผ์™€ ์ƒ๋ช…๊ณผํ•™ ์˜์—ญ์œผ๋กœ ๋ณธ๊ฒฉ ํ™•์žฅํ•˜๊ณ  ์žˆ์–ด์š”. ๋ฏธ์‹œ๊ฐ„ ๋ฐ์ดํ„ฐ์„ผํ„ฐ ์ฐฉ๊ณต๋ถ€ํ„ฐ ์งˆ๋ณ‘ ์น˜๋ฃŒ๋ฅผ ์œ„ํ•œ ์žฌ๋‹จ ์„ค๋ฆฝ๊นŒ์ง€, ์ตœ๊ทผ ํ–‰๋ณด๋Š” “AI๋ฅผ ์–ด๋””๊นŒ์ง€ ์“ธ ๊ฒƒ์ธ๊ฐ€”๋ผ๋Š” ์งˆ๋ฌธ์— ๋Œ€ํ•œ ๊ตฌ์ฒด์ ์ธ ๋‹ต์„ ๋ณด์—ฌ์ฃผ๊ณ  ์žˆ๋‹ต๋‹ˆ๋‹ค.

๋ฌผ๋ฆฌ์  ๊ธฐ๋ฐ˜๊ณผ ์‚ฌํšŒ์  ์ฑ…์ž„, ๋™์‹œ์— ์Œ“์•„ ์˜ฌ๋ฆฌ๋‹ค

The first steel beams went up this week at our Michigan Stargate site with Oracle and Related Digital

[1]

๋ฏธ์‹œ๊ฐ„ Stargate ๋ถ€์ง€์—์„œ ์˜ค๋ผํด, Related Digital๊ณผ ํ•จ๊ป˜ ์ฒซ ์ฒ ๊ณจ ๊ตฌ์กฐ๋ฌผ์ด ์˜ฌ๋ผ๊ฐ”์–ด์š”. ๋Œ€๊ทœ๋ชจ AI ์ธํ”„๋ผ๊ฐ€ ๋ง ๊ทธ๋Œ€๋กœ ๋•… ์œ„์— ์„ธ์›Œ์ง€๊ธฐ ์‹œ์ž‘ํ•œ ์ˆœ๊ฐ„์ด์—์š”.

๊ฐ™์€ ๋‚ , ์ด์™€ ๋งž๋ฌผ๋ ค OpenAI ์žฌ๋‹จ์˜ ๋ฐฉํ–ฅ๋„ ๊ณต๊ฐœ๋์–ด์š”. ์žฌ๋‹จ์€ ์ฒซ 1๋…„๊ฐ„ ์ตœ์†Œ 10์–ต ๋‹ฌ๋Ÿฌ๋ฅผ ํˆฌ์žํ•˜๋ฉฐ ์‹ ์ข… ๋ฐ”์ด์˜ค ์œ„ํ˜‘, ๊ฒฝ์ œ ๊ตฌ์กฐ์˜ ๊ธ‰๊ฒฉํ•œ ๋ณ€ํ™”, AI์˜ ๋ณตํ•ฉ์  ์‚ฌํšŒ ์˜ํ–ฅ ๋“ฑ์„ ํ•ต์‹ฌ ๊ณผ์ œ๋กœ ์‚ผ๋Š”๋‹ค๊ณ  ํ•ด์š”.

AI will help discover new science, such as cures for diseases, which is perhaps the most important way to increase quality of life long-term. AI will also present new threats to society that we have to address. No company can sufficiently mitigate these on their own; we will need a society-wide response.

[2]

AI๊ฐ€ ๊ฐ€์ ธ์˜ฌ ํ˜œํƒ๊ณผ ์œ„ํ˜‘์„ ๋™์‹œ์— ์ธ์ •ํ•˜๋ฉด์„œ, ์–ด๋А ํ•œ ๊ธฐ์—…์ด ํ˜ผ์ž ํ•ด๊ฒฐํ•  ์ˆ˜ ์—†๋‹ค๊ณ  ๋ถ„๋ช…ํžˆ ์„ ์„ ๊ทธ์—ˆ์–ด์š”. ‘AI ์•ˆ์ „’์„ ‘ํšŒ๋ณต ํƒ„๋ ฅ์„ฑ(Resilience)’ ๊ด€์ ์œผ๋กœ ์žฌ์ •์˜ํ•˜๊ฒ ๋‹ค๋Š” ๊ฒƒ๋„ ์ฃผ๋ชฉํ•  ๋งŒํ•œ ์‹œ๊ฐ์ด์—์š”.

โ†ป ๊ณต์œ ํ•œ ๊ธ€

RT @nanransohoff: The new OpenAI nonprofit just announced that it aims to spend $1B in its *first year* and will be led by two superb humansโ€ฆ

[3]

์ด ์†Œ์‹์— ์ฃผ๋ชฉํ•œ ์™ธ๋ถ€ ๊ด€๊ณ„์ž์˜ ๊ธ€๋„ ๊ณต์œ ํ–ˆ์–ด์š”. ์žฌ๋‹จ ์ถœ๋ฒ”์— ๋Œ€ํ•œ ์—…๊ณ„์˜ ๋ฐ˜์‘์ด ๋น ๋ฅด๊ฒŒ ํ™•์‚ฐ๋˜๊ณ  ์žˆ๋‹ค๋Š” ๊ฑธ ๊ฐ„์ ‘์ ์œผ๋กœ ๋“œ๋Ÿฌ๋‚ธ ์…ˆ์ด์—์š”.

๊ฐœ์ธ์ด ์—ฐ๊ตฌ์†Œ๊ฐ€ ๋˜๋Š” ์‹œ๋Œ€, AI์˜ ์ง„์งœ ๊ฐ€๋Šฅ์„ฑ

“The chat bots empowered me as an individual to act with the power of a research institute – planning, education, troubleshooting, compliance, and yes, real scientific design work in converting genomic data to a vaccine prescription and designing the treatment protocol around it. But they worked alongside humans at every step. The combination is what made it possible.”

[4]

ChatGPT์™€ LLM(๊ฑฐ๋Œ€ ์–ธ์–ด ๋ชจ๋ธ)์„ ํ™œ์šฉํ•ด ๋ฐ˜๋ ค๊ฒฌ์„ ์œ„ํ•œ mRNA ๋ฐฑ์‹  ํ”„๋กœํ† ์ฝœ์„ ์ง์ ‘ ์„ค๊ณ„ํ•œ Paul์˜ ์ด์•ผ๊ธฐ๋ฅผ ์†Œ๊ฐœํ–ˆ์–ด์š”. ๊ฐœ์ธ์ด AI๋ฅผ ๋„๊ตฌ ์‚ผ์•„ ์—ฐ๊ตฌ์†Œ ์ˆ˜์ค€์˜ ์ž‘์—…์„ ํ•ด๋‚ธ ์‹ค์ œ ์‚ฌ๋ก€์˜ˆ์š”.

“์ด๊ฑด ๋ฐ”๋กœ ํšŒ์‚ฌ๊ฐ€ ๋˜์–ด์•ผ ํ•œ๋‹ค”๋Š” ๋ฐ˜์‘์—์„œ, AI ๊ฐœ์ธ ์ž„ํŒŒ์›Œ๋จผํŠธ(๊ฐœ์ธ ์—ญ๋Ÿ‰ ๊ฐ•ํ™”)๊ฐ€ ์ƒˆ๋กœ์šด ์‚ฐ์—…์„ ๋งŒ๋“ค ์ˆ˜ ์žˆ๋‹ค๋Š” ์‹œ๊ฐ์ด ์ž˜ ๋“œ๋Ÿฌ๋‚˜์š”. ๋‹ค๋งŒ “์‰ฌ์›Œ์•ผ ํ•  ์ผ์ธ๋ฐ ์•„์ง์€ ์•„๋‹ˆ๋‹ค”๋ผ๋Š” ๋ง๋„ ํ•จ๊ป˜ ๋ง๋ถ™์˜€์–ด์š”.

์ธํ”„๋ผ๋ฅผ ์Œ“๊ณ , ์žฌ๋‹จ์„ ์„ธ์šฐ๊ณ , ๊ฐœ์ธ์˜ ๊ฐ€๋Šฅ์„ฑ์„ ๋ฐœ๊ฒฌํ•˜๋Š” ํ๋ฆ„์ด ๋™์‹œ์— ์ง„ํ–‰๋˜๊ณ  ์žˆ์–ด์š”. AI๊ฐ€ ๊ธฐ์ˆ ์„ ๋„˜์–ด ์‚ฌํšŒ ๊ตฌ์กฐ ์ž์ฒด๋ฅผ ๋ฐ”๊พธ๋Š” ๋‹จ๊ณ„์— ์ง„์ž…ํ•˜๊ณ  ์žˆ๋‹ค๋Š” ์‹ ํ˜ธ๋กœ ์ฝํžŒ๋‹ต๋‹ˆ๋‹ค.

๐Ÿ“š ์ถœ์ฒ˜ (Sources)

  1. [1]@sama on ๐• ยท 2026-03-27 โ€” “The first steel beams went up this week at our Michigan Stargate site with Oracle and Relaโ€ฆ”
  2. [2]@sama on ๐• ยท 2026-03-24 โ€” “AI will help discover new science, such as cures for diseases, which is perhaps the most iโ€ฆ”
  3. [3]โ†ป RT@sama on ๐• ยท 2026-03-24 โ€” “The new OpenAI nonprofit just announced that it aims to spend $1B in its *first year” and โ€ฆ”
  4. [4]@sama on ๐• ยท 2026-03-27 โ€” “The coolest meeting I had this week with was Paul, who used ChatGPT and other LLMs to creaโ€ฆ”
โ† Back to FanPage
Scroll to Top