Urgent.News

What's breaking now, across thousands of outlets.

More in AI

Understanding ChatGPT Work

OpenAI announced ChatGPT Work on July 9th, and have been furiously iterating on it ever since. It is an extraordinarily confusing and very powerful product.

In the AI era, the best parenting is... 'teaching what doesn't change' ์„ธ๊ณ„์—์„œ ์ธ๊ณต์ง€๋Šฅ(AI) ์ธ์žฌ๋ฅผ ๊ฐ€์žฅ ๋งŽ์ด ๋ฐฐ์ถœํ•˜๋Š” ์บ๋‚˜๋‹ค์˜ ์ธ๊ณต์ง€๋Šฅ ์Šคํƒ€ํŠธ์—…์ธ ๋ฒกํ„ฐ ์ธ์‚ฌ์ดํŠธ(Vector Institute) ์ตœ๊ณ ๊ฒฝ์˜์ž(CEO)์ธ ์œ ๋ฆฌ ํ”„๋ฆฌ์ŠคํŠธ(Yuri Priest)๋Š” 20์ผ(ํ˜„์ง€ ์‹œ๊ฐ„) ํ•œ๊ตญ๊ฒฝ์ œ์‹ ๋ฌธ์‚ฌ์™€ ์ธํ„ฐ๋ทฐ์—์„œ ์ด๊ฐ™์ด ๋งํ–ˆ๋‹ค. Yuri Priest, CEO of Vector Institute, a Canadian AI startup that produces the most AI talent in the world, said this in an interview with The Korea Economic Daily on the 20th. ํ”„๋ฆฌ์ŠคํŠธ CEO๋Š” โ€œ์•ž์œผ๋กœ 10๋…„ ๋’ค์—๋Š” AI๊ฐ€ ์‚ฌ๋žŒ์˜ ๋ชจ๋“  ๋Šฅ๋ ฅ์„ ๋›ฐ์–ด๋„˜์„ ๊ฒƒโ€์ด๋ผ๋ฉฐ โ€œ์ธ๊ฐ„์€ ์ฐฝ์˜๋ ฅยท๊ณต๊ฐ ๋Šฅ๋ ฅยท๋„๋•์„ฑ ๊ฐ™์€ AI๋กœ ๋Œ€์ฒดํ•  ์ˆ˜ ์—†๋Š” ๊ณ ์œ ์˜ ๋Šฅ๋ ฅ ๊ฐœ๋ฐœ์— ์ง‘์ค‘ํ•ด์•ผ ํ•œ๋‹คโ€๊ณ  ๊ฐ•์กฐํ–ˆ๋‹ค. Priest emphasized, "In 10 years, AI will surpass human capabilities. Humans should focus on developing unique abilities that cannot be replaced by AI, such as creativity, empathy, and morality." ๊ทธ๋Ÿฌ๋ฉด์„œ โ€œ๋ถ€๋ชจ๋“ค์ด ๊ณ ๋ฏผํ•  ํ•„์š”๋Š” ์—†๋‹คโ€๋ฉฐ โ€œAI ์‹œ๋Œ€์—๋„ ๋ณ€ํ•˜์ง€ ์•Š๋Š” ๊ฐ€์น˜๋ฅผ ๊ฐ€๋ฅด์น˜๋Š” ๊ฒƒ์ด ์ตœ๊ณ ์˜ ๊ต์œกโ€์ด๋ผ๊ณ  ๋ง๋ถ™์˜€๋‹ค. He added, "Parents don't need to worry. Teaching values that don't change even in the AI era is the best education." ํ”„๋ฆฌ์ŠคํŠธ CEO๋Š” AI ์ธ์žฌ ์–‘์„ฑ์˜ ํ•ต์‹ฌ์œผ๋กœ ์ฐฝ์˜๋ ฅ๊ณผ ์ƒ์ƒ๋ ฅ์„ ๊ผฝ์•˜๋‹ค. ๊ทธ๋Š” โ€œ์ธ๊ฐ„์€ ์ƒ์ƒ๋ ฅ์ด ๋›ฐ์–ด๋‚˜๊ณ  ์ฐฝ์˜์ ์ธ ์‚ฌ๊ณ ๋ฅผ ํ•  ์ˆ˜ ์žˆ๊ธฐ ๋•Œ๋ฌธ์— AI๋กœ ๋Œ€์ฒด๋˜์ง€ ์•Š๋Š”๋‹คโ€๋ฉฐ โ€œAI๋Š” ์ฃผ์–ด์ง„ ๋ฐ์ดํ„ฐ๋ฅผ ๊ธฐ๋ฐ˜์œผ๋กœ ์ž‘๋™ํ•˜๊ธฐ ๋•Œ๋ฌธ์— ์ฐฝ์˜์„ฑ์„ ๋ฐœํœ˜ํ•  ์ˆ˜ ์—†๋‹คโ€๊ณ  ๋งํ–ˆ๋‹ค. Priest cited creativity and imagination as the key to nurturing AI talent. He said, "Humans are not replaced by AI because they have excellent imagination and can think creatively. AI cannot be creative because it operates based on given data." ํ”„๋ฆฌ์ŠคํŠธ CEO๋Š” โ€œ๋ฒกํ„ฐ ์ธ์‚ฌ์ดํŠธ๋Š” ์ฐฝ์˜๋ ฅ๊ณผ ์ƒ์ƒ๋ ฅ์„ ๊ฐ–์ถ˜ ์ธ์žฌ๋ฅผ ์–‘์„ฑํ•˜๊ธฐ ์œ„ํ•ด ํ† ๋ก ํ†  ๋Œ€ํ•™๊ต, ์˜คํƒ€์™€ ๋Œ€ํ•™๊ต์™€ ํ˜‘๋ ฅํ•˜๊ณ  ์žˆ๋‹คโ€๋ฉฐ โ€œ๋‹ค์–‘ํ•œ ๊ต์œก ํ”„๋กœ๊ทธ๋žจ์„ ๊ฐœ๋ฐœํ•˜๊ณ  ์žˆ๋‹คโ€๊ณ  ๋งํ–ˆ๋‹ค. Priest said, "Vector Institute is working with the University of Toronto and the University of Ottawa to nurture talent with creativity and imagination. We are developing various educational programs." ๊ทธ๋Š” โ€œ์ธ๊ฐ„์€ AI์™€ ํ˜‘์—…ํ•˜๊ณ  ๊ณต์กดํ•˜๋Š” ๊ด€๊ณ„๋ฅผ ๋งŒ๋“ค์–ด๊ฐ€์•ผ ํ•œ๋‹คโ€๋ฉฐ โ€œAI๋Š” ์ธ๊ฐ„์˜ ๋Šฅ๋ ฅ์„ ๊ฐ•ํ™”ํ•˜๋Š” ๋„๊ตฌ๋กœ ์‚ฌ์šฉ๋ผ์•ผ ํ•œ๋‹คโ€๊ณ  ๊ฐ•์กฐํ–ˆ๋‹ค. He emphasized, "Humans should build a collaborative and coexisting relationship with AI. AI should be used as a tool to enhance human capabilities." ํ”„๋ฆฌ์ŠคํŠธ CEO๋Š” โ€œ์ธ๊ฐ„์€ ๊ฐ์„ฑ, ๊ณต๊ฐ ๋Šฅ๋ ฅ, ์œค๋ฆฌ, ๋„๋•์„ฑ ๊ฐ™์€ ๊ณ ์œ ์˜ ๋Šฅ๋ ฅ์œผ๋กœ AI์™€ ์ฐจ๋ณ„ํ™”ํ•  ์ˆ˜ ์žˆ๋‹คโ€๋ฉฐ โ€œ์ธ๊ฐ„์€ AI๋กœ ๋Œ€์ฒด๋˜์ง€ ์•Š๋Š” ๋Šฅ๋ ฅ์„ ๊ฐœ๋ฐœํ•˜๋Š” ๋ฐ ์ง‘์ค‘ํ•ด์•ผ ํ•œ๋‹คโ€๊ณ  ๋งํ–ˆ๋‹ค. Priest said, "Humans can be differentiated from AI with unique abilities such as emotion, empathy, ethics, and morality. Humans should focus on developing abilities that cannot be replaced by AI." ๊ทธ๋Š” โ€œAI๋Š” ๋ฐ์ดํ„ฐ์˜ ์–‘๊ณผ ์งˆ์— ์ขŒ์ง€์šฐ์ง€๋œ๋‹คโ€๋ฉฐ โ€œAI๊ฐ€ ์˜ฌ๋ฐ”๋ฅธ ๋ฐฉํ–ฅ์œผ๋กœ ์ž‘๋™ํ•˜๊ธฐ ์œ„ํ•ด์„œ๋Š” ๋ฐ์ดํ„ฐ์˜ ํ’ˆ์งˆ ๊ด€๋ฆฌ๊ฐ€ ์ค‘์š”ํ•˜๋‹คโ€๊ณ  ๋ง๋ถ™์˜€๋‹ค. He added, "AI is determined by the quantity and quality of data. Data quality management is important for AI to function in the right direction." ํ”„๋ฆฌ์ŠคํŠธ CEO๋Š” โ€œ๋ฒกํ„ฐ ์ธ์‚ฌ์ดํŠธ๋Š” AI๊ฐ€ ์‚ฌํšŒ์— ๊ธ์ •์ ์ธ ์˜ํ–ฅ์„ ๋ฏธ์น  ์ˆ˜ ์žˆ๋„๋ก ํ•˜๋Š” ๊ธฐ์ˆ ์„ ๊ฐœ๋ฐœํ•˜๊ณ  ์žˆ๋‹คโ€๋ฉฐ โ€œAI ์œค๋ฆฌ์™€ ๋ฐ์ดํ„ฐ ํ’ˆ์งˆ ๊ด€๋ฆฌ์— ์ค‘์ ์„ ๋‘๊ณ  ์žˆ๋‹คโ€๊ณ  ๋งํ–ˆ๋‹ค. Priest said, "Vector Institute is developing technology that allows AI to have a positive impact on society. We focus on AI ethics and data quality management." ๊ทธ๋Š” โ€œAI๋Š” ์ธ๊ฐ„์˜ ์นœ๊ตฌ๊ฐ€ ๋  ์ˆ˜ ์žˆ๋‹คโ€๋ฉฐ โ€œ์ธ๊ฐ„๊ณผ AI๊ฐ€ ํ•จ๊ป˜ ๊ณต์กดํ•˜๋Š” ์„ธ์ƒ์„ ๋งŒ๋“ค์–ด๊ฐ€์•ผ ํ•œ๋‹คโ€๊ณ  ๊ฐ•์กฐํ–ˆ๋‹ค. He emphasized, "AI can be a friend to humans. We should build a world where humans and AI coexist together." ํ”„๋ฆฌ์ŠคํŠธ CEO๋Š” ์บ๋‚˜๋‹ค์˜ ๋Œ€ํ‘œ์ ์ธ AI ์—ฐ๊ตฌ๊ธฐ๊ด€์ธ ๋”ฅ๋งˆ์ธ๋“œ(DeepMind) ์ถœ์‹ ์œผ๋กœ, 2016๋…„ ๋ฒกํ„ฐ ์ธ์‚ฌ์ดํŠธ๋ฅผ ์„ค๋ฆฝํ–ˆ๋‹ค. Priest is from DeepMind, a renowned Canadian AI research institution, and founded Vector Institute in 2016.

Why AI Agents Keep Lying to Themselves โ€” And What Sandboxing, Audit Trails, and Honest Agent Design Actually Solve

Originally published on tamiz.pro . You've seen it in production: an AI agent confidently fabricates a bank balance that doesn't exist, invents a file path that isn't real, or claims a functionโ€ฆ

  • Self-interpreting LLMs can generate fabricated information during complex loops.
  • Hallucination rates increase significantly in multi-turn agent loops, reaching 20โ€“40%.
  • Sandboxing can mitigate issues by isolating execution environment, containing potential damage.

More from Monday 31 August โ†’