Continuous Learning_Startup & Investment
2.44K subscribers
513 photos
5 videos
16 files
2.74K links
We journey together through the captivating realms of entrepreneurship, investment, life, and technology. This is my chronicle of exploration, where I capture and share the lessons that shape our world. Join us and let's never stop learning!
Download Telegram
This is the night that this phrase comes to mind.

1. Your friends can help you see what you cannot see.
2. I will help them and they help you and we can do better and go further.
3. I have hundreds of friends, mentors.
4. Friends let you help them. Friends had a full trust and faith to me via conversation which make me I am a worthy person.

1. Make friendship with high priority
2. Make concious good friendship. Make a rituals. Talk about what friendship is and how to improve this.
3. Share what you want and your dream to friends.
4. Friends tell you not what you want but what I need to do.
5. Life is most fullfilling teamsports
๐Ÿ‘2
์ตœ์ธ์•„ ๋Œ€ํ‘œ๋‹˜์˜ '๋‚ด๊ฐ€ ๊ฐ€์ง„ ๊ฒƒ์„ ์„ธ์ƒ์ด ์›ํ•˜๊ฒŒ ํ•˜๋ผ' ์ฑ…์— ์ด๋Ÿฐ ๊ตฌ์ ˆ์ด ๋‚˜์˜จ๋‹ค. 2007๋…„ ํ•œ ์‹ ๋ฌธ์— ๊ธฐ๊ณ ํ–ˆ๋˜ ๊ธ€์˜ ๋‚ด์šฉ์ด๋ผ๊ณ .
"์ Š๊ธฐ๋Š” ์‰ฝ๋‹ค. ๋ชจ๋‘ ์ Š๋‹ค, ์ฒ˜์Œ์—”. ๋Š™๊ธฐ๋Š” ์‰ฝ์ง€ ์•Š๋‹ค. ์„ธ์›”์ด ๊ฑธ๋ฆฐ๋‹ค. ์ Š์Œ์€ ์ฃผ์–ด์ง„๋‹ค. ๋Š™์Œ์€ ์ด๋ฃจ์–ด์ง„๋‹ค. ๋Š™๊ธฐ ์œ„ํ•ด์„  ์„ธ์›”์— ์„ž์„ ๋งˆ๋ฒ•์„ ๋งŒ๋“ค์–ด๋‚ด์•ผ ํ•œ๋‹ค."
์ด ๊ตฌ์ ˆ์„ ๋ณด๊ณ  ์˜ˆ์ „์— ์ฝ์—ˆ๋˜ ํ•œ ๋ฌธ์žฅ์ด ๋– ์˜ฌ๋ž๋‹ค. '์ Š๋‹ค'๋Š” ํ˜•์šฉ์‚ฌ์ธ๋ฐ, '๋Š™๋‹ค'๋Š” ๋™์‚ฌ๋ผ๋Š” ๋ฌธ์žฅ. ๊ทธ๋ ‡๊ธฐ์— ์‚ฌ๋žŒ๋งˆ๋‹ค ์–ด๋–ป๊ฒŒ(์–ด๋–ค ๋ชจ์Šต์œผ๋กœ) ๋Š™์–ด๊ฐˆ์ง€ ๊ฐœ์ธ์ด ์„ ํƒํ•  ์ˆ˜ ์žˆ๊ณ , ๊ทธ ์„ ํƒ์— ๋”ฐ๋ผ ๋งŒ๋“ค์–ด ๊ฐ„ ๊ฒƒ์ด ์ž์‹ ์˜ '๋Š™์€ ๋ชจ์Šต'์ด๋ผ๋Š” ๊ฒƒ.
์ž์‹ ์˜ '๋Š™์Œ'์ด '๋Š™์–ด๋ฒ„๋ ธ๋‹ค'๋ผ๋Š” ๋‹จ์–ด๋กœ ํ‘œํ˜„๋˜๋Š” ๊ฒƒ์ด ์•„๋‹ˆ๋ผ, ์–ด๋–ค ๋ชจ์Šต์œผ๋กœ ๋Š™์–ด๊ฐ€๊ณ  ์‹ถ์€์ง€์— ๋Œ€ํ•ด ์ƒ๊ฐํ•˜๊ณ  ๊ทธ๊ฒƒ์„ ์ ๊ทน์ ์œผ๋กœ ์ด๋ฃจ๋ ค๊ณ  ํ•œ ๊ฒฐ๊ณผ๋กœ์จ '๋Š™์Œ'์„ ๋ฐ”๋ผ๋ณผ ์ˆ˜ ์žˆ๋‹ค๋ฉด, ๊ฑฑ์ •์˜ ๋Œ€์ƒ์ด ์•„๋‹ˆ๋ผ ์„ค๋ ˜์˜ ๋Œ€์ƒ์œผ๋กœ ๋Š™์Œ์„ ๋ฐ”๋ผ๋ณผ ์ˆ˜๋„ ์žˆ๊ฒ ๋‹ค๋Š” ์ƒ๊ฐ์ด ๋“ค์—ˆ๋‹ค.
๊ทธ๋Ÿฐ ์˜๋ฏธ์—์„œ ์ด ์ฑ…์€, ๋ˆ„๊ตฐ๊ฐ€์—๊ฒ ์ž๊ธฐ ๊ณ„๋ฐœ์„œ๊ฒ ์ง€๋งŒ, ์ž์‹ ์„ ์ถฉ๋ถ„ํžˆ ์‚ฌ๋ž‘ํ•˜๋Š” ์‚ฌ๋žŒ์ด ๋˜์ž๋Š” ๊ฒƒ๊ณผ ๊ทธ๋ ‡๊ฒŒ ํ•˜๊ธฐ ์œ„ํ•ด ๊ฐ€์ ธ์•ผ ํ•˜๋Š” ํƒœ๋„๋ฅผ ์–˜๊ธฐํ•˜๋Š” ์‚ถ์˜ ์ฒ ํ•™์— ๋Œ€ํ•œ ์ด์•ผ๊ธฐ์ด๋‹ค.
๐Ÿ‘3
Picking Teams in AI

Yesterday, Databricks announced its intent to acquire Mosaic for $1.3b. Perhaps not coincidentally, Snowflake announced a deepened partnership with Nvidia to offer customers models & training on Nvidiaโ€™s Nemo platform.
Clouds are picking teams in one of the most important dislocations in software.
Cloud <- -> LLM Infrastructure
Microsoft <- -> OpenAI
Snowflake <- -> Nvidia
Databricks <- -> Mosaic
Google <- -> Anthropic
Oracle <- -> Cohere
Amazon <- -> HuggingFace
Microsoft has invested over $10b, plus significant development efforts to work with OpenAI. In addition, Microsoft & Snowflake announced a deeper AI go-to-market partnership with Snowflake.
Snowflakeโ€™s partnership with Nvidia positions Snowflakeโ€™s cloud as a broader infrastructure platform.
DataBricks, whose business revolves around Spark operating in customer environments, has announced plans to acquire Mosaic, a vertically integrated model training & management system that functions on similar workloads.
Google has invested hundreds of millions into Anthropic, complementing its efforts with Google Brain.
Oracle has paralleled Microsoftโ€™s OpenAI partnership with Cohere, investing & seeking to build a product for Oracleโ€™s cloud with Cohere.
Amazon has announced HuggingFace LLMs on their Sagemaker product, embracing the open-source community.
Cloudflare, which has seen tremendous interest in its R2 storage product for model training because of a lower cost storage infrastructure, had partnered with Mosaic. Itโ€™s unclear how the Databricks acquisition might change that relationship.
Cloud infrastructure players are picking teams within the infrastructure layer. Most major cloud players have picked an LLM partner & perhaps will choose multiple.
For startups building LLM-based applications & infrastructure, this alters the calculus of selecting a cloud. Five years ago, many startups defaulted to AWS for the generous credits, broad catalog, & rapid pace of innovation.
LLM-enabled apps require customer data to train, propelling data security to the top of the list for most enterprise buyers. Startups may begin to pick clouds to reach a particular class of customer, the security promises the underlying platforms provide, & then available algorithms & cost.
Access to particular models may be a consideration, but given the rapid advances in open-source that advantage will likely erode over time.

https://tomtunguz.com/ai-consolidation/
๐Ÿ‘1
Economic Uncertainty Elevates Importance of Machine Learning Observability Arize
Forwarded from Notes from Sigrid Jin (Sigrid Jin)
https://news.hada.io/topic?id=9520&utm_source=discord&utm_medium=bot&utm_campaign=4
LLM์˜ ๋“ฑ์žฅ์œผ๋กœ Vector ํ˜•์‹์˜ ๋ฐ์ดํ„ฐ ์ˆ˜์†ก์‹ ์ด JSON์ฒ˜๋Ÿผ ๋Œ€์ค‘ํ™”๋  ๊ฒƒ์ด๋ผ๋Š” ์ฃผ์žฅ
I think weโ€™re starting to see frothy valuations coming back in certain areas, especially genAI. Weโ€™re talking about $30mm pre for pre-seed, pre-product teams. I donโ€™t think thatโ€™s particularly healthy.

If you were lucky enough to raise big venture money, one thing I tell founders is treat the money as if it was a repayable loan.

No one throws a block party for their mortgage approval and part of the reason is they know too well that the money needs to be repaid at some point. Itโ€™s the same for VC funding too - except the investor expectation/hope is only higher (everybody wants a 10x return). All kinds of problems happen when founders think the VC money is โ€œtheirsโ€.

Chang Kim
โ€œApproved H1-B holders as of July 16, 2023 will receive an open work permit of up to 3yrs, which means they will be able to work for almost any employer anywhere in Canada.โ€

Canada is finally ready to eat USAโ€™s lunch on poor immigration policy!

As part of Canadaโ€™s first-ever Tech Talent Strategy, Minister Fraser announced the following aggressive attraction measures:
โ€ข the creation of an open work permit stream for H-1B specialty occupation visa holders in the US to apply for a Canadian work permit, and study or work permit options for their accompanying family members
โ€ข the development of an Innovation Stream under to the International Mobility Program to attract highly talented individuals, options for which include
โ—ฆ employer-specific work permits for up to five years for workers destined to work for a company identified by the Government of Canada as contributing to our industrial innovation goals
โ—ฆ open work permits for up to five years for highly skilled workers in select in-demand occupations
โ€ข a return to the 14-day service standard for work permits under the Global Skills Strategy
โ€ข the promotion of Canada as a destination for digital nomads
โ€ข the creation of a STEM-specific draw under category-based selection to issue additional invitations to apply under the Express Entry program
โ€ข improvements to the Start-up Visa Program
โ—ฆ We allocated more spots to this program for 2023, with further increases planned for 2024 and 2025.
โ—ฆ Applicants will be able to apply for a work permit that is up to three years in duration instead of one year.
โ—ฆ Applicants will be able to apply for an open work permit instead of one that limits them to working for their own start-up.
โ—ฆ Weโ€™re making this three-year open work permit available to each member of the entrepreneurial team instead of only those who are essential and urgently needed in Canada.
โ—ฆ Weโ€™re prioritizing applications that are supported by venture capital, angel investor groups and business incubators and have capital committed, along with applications that are supported by business incubators who are members of Canadaโ€™s Tech Network.
1. OpenAI Plans ChatGPT โ€˜Personal Assistant for Work,โ€™ Setting Up Microsoft Rivalry.

โ€ข OpenAI wants to build ChatGPT into a personal assistant
โ€ข The effort could take years
โ€ข It hinges on training AI models on personal data without leaking that information

Earlier this month, Altman made a key hire that could be related to the assistant effort. OpenAI named former Facebook leader Peter Deng as vice president of consumer product. In a LinkedIn post announcing his new role, Deng cited the companyโ€™s desire to develop โ€œassistive capabilities that make people more productive and creative.โ€ Deng was previously the first product manager of Facebook Messenger and later led product at Instagram before serving as a product leader at Uber and then Airtable, a productivity software firm.


2. ChatGPT last fall, OpenAI started charging individuals and companies for an upgraded version, which recently had more than 2 million subscribers and is on pace to generate hundreds of millions of dollars per year

=> Chat GPT paid customer 2m x $20 -> monthly revenue $40m (์›” 520๋ฐฑ์–ต) yearly $480m 6280์–ต)

Open AIโ€™s last year loss $540 million(7000์–ต)

3. To boost ChatGPTโ€™s value, OpenAI is considering opening a type of app store in which its customers would sell customized versions of the chatbot that could help other businesses draft marketing materials, identify fraudulent transactions or build customer service chatbots, The Information reported last week.

Challenges
1. But to operate in a more personalized fashion and respond quickly the way Appleโ€™s Siri does, the prospective OpenAI assistant software may need to be partly stored on usersโ€™ devices.

MSโ€™s situation
Large-language models that allow ChatGPT and other software to understand conversational commands are relatively new, although Microsoft is already charging a 40% premium to Office 365 customers that want to use OpenAIโ€™s LLMs to automate tasks such as creating PowerPoint presentations based on text documents, summarizing meetings or drafting email responses.

New challengers


If open-source LLMs continue to progress, developers may be able to create their own assistants. One example is Personal AI, a startup founded in 2020 that has raised $7.8 million. It has trained open-source modelsโ€”as well as its proprietary LLMโ€”on an individualโ€™s data and made the models small enough to run entirely on laptops or phones.
ํ‰๋ฒ”ํ•œ ์‚ฌ๋žŒ๋“ค์ด ์œ„๋Œ€ํ•œ ๊ฒฐ๊ณผ ๋งŒ๋“ค์–ด๋‚ด๋Š” ์ด์•ผ๊ธฐ๋Š” ์†Œ์„ค๊ฐ™์ง€๋งŒ, ๋งค๋ ฅ์ ์ž…๋‹ˆ๋‹ค. ์˜ํ™”๊ฐ™๊ธฐ๋งŒ ํ•œ ์ด์•ผ๊ธฐ ๊ฐ™์€๋ฐ, ์‹ค์ œ๋กœ ๋ฒ„ํ•์ด ๊ทธ๋Ÿฐ ์‚ฌ๋ก€๋ฅผ ์–˜๊ธฐํ•˜๋‹ˆ ๋” ์šธ๋ฆฌ๊ธฐ๋„ ํ•˜๊ณ ์š”. ์‚ฌ์‹ค ํ‰๋ฒ”ํ•œ ์‚ฌ๋žŒ๋“ค์ด ์œ„๋Œ€ํ•œ ํšŒ์‚ฌ๋ฅผ ๋งŒ๋“ ๋‹ค๋Š” ๋ง ์ž์ฒด๊ฐ€ ๋ชจ์ˆœ์ด์ง€ ์•Š๋‚˜ ์‹ถ๊ธฐ๋„ ํ•ฉ๋‹ˆ๋‹ค. ์œ„๋Œ€ํ•œ ํšŒ์‚ฌ ๋งŒ๋“ค๊ธฐ ์œ„ํ•ด์„œ ํ‰๋ฒ”ํ•œ ๊ฒƒ ์˜ค๋ž˜ ํฌ๊ธฐํ•˜์ง€ ์•Š๊ณ  ์Œ“์•„๊ฐ„ ๊ฒƒ ์ž์ฒด๊ฐ€ ์•ˆ ํ‰๋ฒ”ํ•œ ๊ฒƒ์ด๋‹ˆ๊นŒ์š”. ๋ชจ๋‘๊ฐ€ ์•Œ์ง€๋งŒ, ์ง€ํ‚ค์ง€ ๋ชปํ•˜๋Š” ๊ฒƒ์ด๋‹ˆ๊นŒ์š”. ์‚ฌ์‹ค ์„ฑ๊ณต์ด ๊ฑฐ์ฐฝํ•œ ๊ฒƒ์ธ๊ฐ€ ์‹ถ์Šต๋‹ˆ๋‹ค. ํ•˜๋ฃจํ•˜๋ฃจ ์Œ“์•„ ์˜ฌ๋ฆฌ๋Š” ๊ฒƒ์ธ๋ฐ์š”. ๊ทธ๊ฒƒ์„ ํ•˜๋ฃจ ํ•˜๊ธฐ๋Š” ์‰ฌ์›Œ๋„, 100์ผ, 1๋…„, 10๋…„ ํ•˜๊ธฐ ์–ด๋ ค์šฐ๋‹ˆ๊นŒ ์–ด๋ ค์šด ๊ฒƒ์ด๊ฒ ์ง€์š”. ์˜ค๋Š˜๋„ ์ž๊ธฐ๋งŒ์˜ ๊ธธ ๊ฑธ์œผ์‹œ๋Š” ๋ชจ๋“  ๋ชจ๋“  ๋ถ„๋“ค๊ป˜ ๋ฒ„ํ•์˜ ํ•œ ๋งˆ๋””๊ฐ€ ์œ„๋กœ์™€ ๊ฒฉ๋ ค๊ฐ€ ๋˜๊ธฐ๋ฅผ ๋ฐ”๋ผ๋ฉฐ ์˜ฌ๋ฆฝ๋‹ˆ๋‹ค.

https://youtu.be/gXFXMCAwP-k
โค1
BZCF | ๋น„์ฆˆ๊นŒํŽ˜
ํ‰๋ฒ”ํ•œ ์‚ฌ๋žŒ๋“ค์ด ์œ„๋Œ€ํ•œ ๊ฒฐ๊ณผ ๋งŒ๋“ค์–ด๋‚ด๋Š” ์ด์•ผ๊ธฐ๋Š” ์†Œ์„ค๊ฐ™์ง€๋งŒ, ๋งค๋ ฅ์ ์ž…๋‹ˆ๋‹ค. ์˜ํ™”๊ฐ™๊ธฐ๋งŒ ํ•œ ์ด์•ผ๊ธฐ ๊ฐ™์€๋ฐ, ์‹ค์ œ๋กœ ๋ฒ„ํ•์ด ๊ทธ๋Ÿฐ ์‚ฌ๋ก€๋ฅผ ์–˜๊ธฐํ•˜๋‹ˆ ๋” ์šธ๋ฆฌ๊ธฐ๋„ ํ•˜๊ณ ์š”. ์‚ฌ์‹ค ํ‰๋ฒ”ํ•œ ์‚ฌ๋žŒ๋“ค์ด ์œ„๋Œ€ํ•œ ํšŒ์‚ฌ๋ฅผ ๋งŒ๋“ ๋‹ค๋Š” ๋ง ์ž์ฒด๊ฐ€ ๋ชจ์ˆœ์ด์ง€ ์•Š๋‚˜ ์‹ถ๊ธฐ๋„ ํ•ฉ๋‹ˆ๋‹ค. ์œ„๋Œ€ํ•œ ํšŒ์‚ฌ ๋งŒ๋“ค๊ธฐ ์œ„ํ•ด์„œ ํ‰๋ฒ”ํ•œ ๊ฒƒ ์˜ค๋ž˜ ํฌ๊ธฐํ•˜์ง€ ์•Š๊ณ  ์Œ“์•„๊ฐ„ ๊ฒƒ ์ž์ฒด๊ฐ€ ์•ˆ ํ‰๋ฒ”ํ•œ ๊ฒƒ์ด๋‹ˆ๊นŒ์š”. ๋ชจ๋‘๊ฐ€ ์•Œ์ง€๋งŒ, ์ง€ํ‚ค์ง€ ๋ชปํ•˜๋Š” ๊ฒƒ์ด๋‹ˆ๊นŒ์š”. ์‚ฌ์‹ค ์„ฑ๊ณต์ด ๊ฑฐ์ฐฝํ•œ ๊ฒƒ์ธ๊ฐ€ ์‹ถ์Šต๋‹ˆ๋‹ค. ํ•˜๋ฃจํ•˜๋ฃจ ์Œ“์•„ ์˜ฌ๋ฆฌ๋Š” ๊ฒƒ์ธ๋ฐ์š”.โ€ฆ
This story deeply resonates with me and its significance can be appreciated by everyone.
AI ์ง„๋ณด/๋‰ด์Šค์— ๊ด€ํ•œ ์ •์ฒด๊ฐ

๋Œ์ด์ผœ ๋ณด๋‹ˆ ์ตœ๊ทผ AI ์†Œ์‹์— ๊ด€ํ•œ ํฌ์ŠคํŒ…์„ ๋œธํ•˜๊ฒŒ ํ•˜๊ณ  ์žˆ์—ˆ์Šต๋‹ˆ๋‹ค. ์—ฌ์ „ํžˆ ์†Œ์‹์€ ๋Š์ž„์—†์ด ์Ÿ์•„์ ธ ๋‚˜์˜ค๊ณ  ์žˆ์ง€๋งŒ ์ „์ฒ˜๋Ÿผ ๋ฏผ๊ฐํ•˜๊ฒŒ ๋ฐ˜์‘ํ•˜์ง€ ์•Š๋Š” ์ด์œ ์— ๊ด€ํ•ด ์ƒ๊ฐํ•ด ๋ณด๊ฒŒ ๋๋Š”๋ฐ์š”. ๊ฐœ์ธ์ ์œผ๋กœ ํ”ผ๋กœ๊ฐ์„ ๋А๊ปด์„œ์ผ ์ˆ˜๋„ ์žˆ์ง€๋งŒ ์š”์ฆ˜์€ ๋ฌด์—‡์„ ๋ด๋„ '์™€์šฐ!'๊ฐ€ ์ „์ฒ˜๋Ÿผ ๋งŽ์ง€ ์•Š์•„์„œ ๋•Œ๋ฌธ์ธ ๊ฒƒ ๊ฐ™๊ธฐ๋„ ํ•ฉ๋‹ˆ๋‹ค.

์–ด๋–ค ์†Œ์‹์„ ๋ณด๊ณ ์„  "์ด๊ฑด ์ถฉ๋ถ„ํžˆ ๋‚˜์˜ฌ๋ฒ•ํ•œ ๋‚ด์šฉ์ด๊ตฐ~"ํ•˜๋ฉฐ... ์Šค์Šค๋กœ์˜ ๋ฐ˜์‘์ด ๋ฌด๋ŽŒ์กŒ๋‹จ ์ž๊ฐ์ด ์žˆ์Šต๋‹ˆ๋‹ค. ๊ตฌ๊ธ€ ๋”ฅ๋งˆ์ธ๋“œ์˜ Gemini๊ฐ€ ๋‹ค์Œ์— ๋ณด์—ฌ์ค„ ๊ฒƒ, ๋ฉ€ํ‹ฐ ๋ชจ๋‹ฌ๋„ ๊ธฐ๋Œ€๋˜์ง€๋งŒ 'ํ”Œ๋ž˜๋‹'์ด ๊ณผ์—ฐ ๋ฌด์—‡์ผ์ง€ ์ •๋„์— ํฐ ํ˜ธ๊ธฐ์‹ฌ์ด ์žˆ๋‹ฌ๊นŒ์š”. ( https://www.wired.com/story/google-deepmind-demis-hassabis-chatgpt ) ๋ฉ€ํ‹ฐ ๋ชจ๋‹ฌ์€ ์ด๋ฏธ ํ”Œ๋ผ๋ฐ๊ณ ๋‚˜ GPT-4 ๋ฐ๋ชจ์—์„œ ๊ทธ ํŽธ๋ฆฐ์„ ๋ดค๊ธฐ ๋•Œ๋ฌธ์— ์–ด๋– ํ•œ ๊ฒƒ์ธ์ง€ ์–ด๋А ์ •๋„ ๊ทธ๋ฆผ์ด ๊ทธ๋ ค์ง€์ง€๋งŒ GPT-4๋„ ๊ณ„์‚ฐ ์ž์› ๋ถ€์กฑ ๋•Œ๋ฌธ์— ๋ฉ€ํ‹ฐ ๋ชจ๋‹ฌ ๊ธฐ๋Šฅ์˜ ๋กค ์•„์›ƒ์„ ์•„์ง ๋ชปํ•˜๊ณ  ์žˆ๋‹ค๊ณ  ํ•˜๋‹ˆ, ๊ณผ์—ฐ Gemini์˜ ๋ฉ€ํ‹ฐ ๋ชจ๋‹ฌ ๊ธฐ๋Šฅ์ด ์„ธ์ƒ์— ๋‚˜์™€๋„ ์„œ๋น„์Šค๋กœ ๊ฒฝํ—˜ํ•  ์ˆ˜ ์žˆ์„๋Ÿฐ์ง€๋Š” ๋˜ ๋ชจ๋ฅด๋Š” ์ผ์ž…๋‹ˆ๋‹ค. ๊ทธ๋ž˜์„œ 'ํ”Œ๋ž˜๋‹'์ด ์ข€ ๋” ๊ด€์ „ ํฌ์ธํŠธ๊ฐ€ ๋  ๊ฒƒ ๊ฐ™์•„์š”. (๊ณผ์—ฐ Monte-Carlo Tree Search + RL์„ Auto-regressive ๋ฐฉ์‹์˜ LLM์— ๊ณ„์‚ฐ ๋น„์šฉ์„ ํฌ๊ฒŒ ๋Š˜๋ฆฌ์ง€ ์•Š๊ณ  ๋ญ”๊ฐ€ ํ˜์‹ ์ ์œผ๋กœ ์ ‘๋ชฉํ•˜๋Š” ๋ฐฉ์‹์ผ๋Ÿฐ์ง€์š”?)

์˜คํ”ˆ ์†Œ์Šค ์ง„์˜์˜ LLM์˜ ๊ฒฝ๋Ÿ‰ํ™”๋‚˜ ์˜ค๋Š˜๋„ ๋Š์ด์ง€ ์•Š๋Š” ์†Œ์‹๋“ค(์˜ค๋Š˜์€ MosaicML์ด ์ธ์ˆ˜๋œ ์†Œ์‹์ด ์žˆ์—ˆ์ฃ )์ด ์žˆ์ง€๋งŒ, GPT-4์™€์˜ ๊ฐ„๊ทน์„ ์ค„์ด๋Š” ๊ฒƒ์ด ๋ชฉํ‘œ๋กœ ๋ณด์—ฌ์„œ ๊ทธ ๊ฐ„๊ทน์ด ์ค„์–ด๋“ค์ง€์–ธ์ • ๊ทธ ์ด์ƒ์˜ ์„ฑ๋Šฅ์€ ๋ณด์—ฌ์ฃผ๊ธฐ ์–ด๋ ค์šธ ๊ฒƒ์ด๋ž€ ์ƒ๊ฐ์ด ๋“ค์–ด์„œ ๋ง‰ ํฅ๋ถ„๋˜๋ฉฐ ๊ธฐ๋Œ€๊ฐ€ ๋˜์ง„ ์•Š๋”๋ผ๊ณ ์š”. GPT-4์˜ ํ™œ์šฉ์—์„œ ์ด๋ฏธ ์–ด๋А ์ •๋„ ํ•œ๊ณ„๋ฅผ ๋งŒ๋‚ฌ๋‹ค๋ฉด ๊ทธ ์ดํ•˜์˜ ๋ชจ๋ธ์—์„œ ๊ฑฐ๊ธธ ๋„˜์–ด์„œ๋Š” ๊ฒƒ์€ ๋ถˆ๊ฐ€๋Šฅํ•˜๊ฒ ์ฃ .

GPT-4๊ฐ€ ๋น„๊ต์  ์ตœ๊ทผ์˜ ๋ฃจ๋จธ๋Œ€๋กœ 220B ๋ชจ๋ธ 8๊ฐœ์˜ Mixture of Experts ํ˜•ํƒœ๋ผ๋ฉด( https://thealgorithmicbridge.substack.com/p/gpt-4s-secret-has-been-revealed ), ์ด ๋˜ํ•œ ๊ฒฝ์ฒœ๋™์ง€ํ•  ํ˜์‹ ์€ ์•„๋‹ˆ๋ผ... ์ƒ˜ ์˜ฌํŠธ๋งŒ์ด ํŠธ์œ—์—์„œ GPT-4๊ฐ€ ๋‚˜์˜ค๊ธฐ ์ „์— ๋ˆ„๋ˆ„ํžˆ ๋งํ•œ๋Œ€๋กœ ๋„ˆ๋ฌด ํฐ ๊ธฐ๋Œ€๋ฅผ ํ•˜๋ฉด ๋ฐ˜๋“œ์‹œ ์‹ค๋งํ•  ๊ฒƒ์ด๋‹ค๋ž€ ๋ง๊ณผ ์•ž๋’ค๊ฐ€ ๋งž๋Š” ์…ˆ์ž…๋‹ˆ๋‹ค. ๋‹น๋ถ„๊ฐ„ OpenAI์˜ ๋กœ๋“œ๋งต์ด GPT-4์˜ ์ดˆ๊ธฐ ์„ฑ๋Šฅ์„ ๋„ˆํ”„์‹œํ‚ค๋ฉด์„œ ๊ณ„์‚ฐ ๋น„์šฉ์„ ์ค„์ด๊ณ  ์†๋„๋ฅผ ๋†’์ด๊ณ  ํ•˜๋Š” ๊ฒƒ์ด ๊ฐ€๋Šฅํ•ด์ง€๋Š” ๊ฒƒ๊ณผ ์ด ๋ฃจ๋จธ๋„ ์•ž๋’ค๊ฐ€ ์ž˜ ๋งž๋Š” ํŽธ์ธ ๋“ฏ ํ•˜๊ณ ์š”.

๊ทธ๋ž˜์„œ ๋‹น๋ถ„๊ฐ„ (๊ฒฝํ•จ ๊ฐ€๋Šฅํ•œ) ์ปคํŒ… ์—ฃ์ง€์—์„œ์˜ ๋งˆ์ง€๋…ธ์„ ์€ GPT-4์ผ ๊ฒƒ์ด๊ณ , ์•ž์„œ ๋งํ•œ ๊ฒƒ ์ฒ˜๋Ÿผ GPT-4๋กœ๋„ ๋‹ฌ์„ฑํ•˜๊ธฐ ์–ด๋ ค์šด ์„œ๋น„์Šค๋Š” ๊ทธ ์ดํ•˜์—์„œ๋„ ์•ˆ๋  ๊ฐ€๋Šฅ์„ฑ์ด ๋†’๋‹ค๋Š” ์˜๋ฏธ๊ฒ ์ฃ . ์•„๋ฌด๋ฆฌ ๊ฐ™์€/๋‹ค๋ฅธ ์—ฌ๋Ÿฌ ๋ชจ๋ธ๋กœ ์•™์ƒ๋ธ”์„ ํ•˜๋˜ ์ฒด์ด๋‹์„ ํ•˜๋˜ ๋ชจ๋ธ๋“ค์˜ ์‚ฌํšŒ๋กœ ์ˆ™์˜ํ•˜๋Š” ์‹œ์Šคํ…œ2๋ฅผ ๊ตฌํ˜„ํ•ด๋„ ๊ธฐ๋ณธ GPT-4์˜ ์„ฑ๋Šฅ์„ ํฌ๊ฒŒ ๋„˜์ง€ ๋ชปํ•˜๊ณ  ๊ณ ๋งŒ๊ณ ๋งŒ ํ•  ๊ฒƒ ๊ฐ™์•„์š”.

์ƒ˜ ์˜ฌํŠธ๋งŒ์ด ํˆฌ์–ด ์ค‘์— ๋ฐœ์–ธํ–ˆ๋“ฏ LLM์—๊ฒŒ ๋„๊ตฌ๋ฅผ ์ฅ์–ด์ฃผ๋Š” ํ”Œ๋Ÿฌ๊ทธ์ธ์€ ๋šœ๊ป‘์„ ์—ด์–ด๋ณด๋‹ˆ PMF(ํ”„๋กœ๋•ํŠธ ๋งˆ์ผ“ ํ•)์„ ๋ชป์ฐพ๋Š” ๋ชจ์–‘์ƒˆ๊ฐ€ ๋์Šต๋‹ˆ๋‹ค. ๊ทธ๋ž˜์„œ OpenAI๋„ ์ž‘์€ ๋ชจ๋ธ์„ ์ง์ ‘ ํ•˜๋ฉฐ ๋ชจ๋ธ ๋งˆ์ผ“์„ ์—ด ๊ฒƒ์ด๋‹ค๋ž€ ์ „๋ง์ด ๋‚˜์˜ค๊ณ  ์žˆ๋Š”๋ฐ์š”. ์ด๊ฒƒ๋„ ๋‚˜์™€๋ด์•ผ ์•Œ๊ฒ ์ง€๋งŒ ์—ญ์‹œ ๋‹น๋ถ„๊ฐ„์˜ ๋งˆ์ง€๋…ธ์„ ์ธ GPT-4์˜ ์„ฑ๋Šฅ ์ดํ•˜๊ฒ ์ฃ .

๋น„์ฆˆ๋‹ˆ์Šค ์ธก๋ฉด์—์„  ๊ธฐ๋ฐ˜์ด ๋˜๋Š” ๊ธฐ์ˆ ์˜ (๋‹น๋ถ„๊ฐ„์˜) ๋ฒ ์ŠคํŠธ ํ”„๋ž™ํ‹ฐ์Šค, ๋ ˆ์‹œํ”ผ๋Š” ๋‚˜์˜จ ์…ˆ์ด๋ผ... ์ €๋Š” ๊ทธ๊ฑธ ํฌ๊ฒŒ ๋‘ ๊ฐ€์ง€๋กœ ๋ดค์Šต๋‹ˆ๋‹ค.

- ๋น„๊ต์  ์ €๋ ดํ•œ: ์ž„๋ฒ ๋”ฉ/๋ฒกํ„ฐ ์„œ์น˜๋ฅผ ๊ทธ๋ผ์šด๋”ฉ์œผ๋กœ โ†’ LLM์˜ ์ปจํ…์ŠคํŠธ ์œˆ๋„์šฐ(4k ~ 16k) ์ฑ„์šฐ๊ธฐ โ†’ ํ…์ŠคํŠธ ์ถœ๋ ฅ ๋˜๋Š” ์ฝ”๋“œ / JSON ๋“ฑ์œผ๋กœ ์ถœ๋ ฅ โ†’ ์ƒŒ๋“œ๋ฐ•์‹ฑ์œผ๋กœ ์ œํ•œํ•˜๊ณ  ๋„๊ตฌ ์—ฐ๋™ ๋˜๋Š” ์ฒด์ด๋‹

- ์ƒ๋Œ€์ ์œผ๋กœ ๋น„์‹ผ: ๊ทธ๋ผ์šด๋”ฉ์œผ๋กœ ์“ธ ์ˆ˜ ์žˆ๋Š” ๋Œ€์šฉ๋Ÿ‰ ํ† ํฐ โ†’ ํฐ ์ปจํ…์ŠคํŠธ ์œˆ๋„์šฐ(32k ~ 100k๊ธ‰)๋ฅผ ๊ฐ€์ง€๋Š” LLM์˜ ์ปจํ…์ŠคํŠธ ์œˆ๋„์šฐ๋ฅผ ์ด ํ† ํฐ์œผ๋กœ ์ฑ„์šฐ๊ธฐ โ†’ ํ…์ŠคํŠธ ์ถœ๋ ฅ ๋˜๋Š” ์ฝ”๋“œ / JSON ๋“ฑ์œผ๋กœ ์ถœ๋ ฅ โ†’ ์ƒŒ๋“œ๋ฐ•์‹ฑ์œผ๋กœ ์ œํ•œํ•˜๊ณ  ๋„๊ตฌ ์—ฐ๋™ ๋˜๋Š” ์ฒด์ด๋‹

๊ทธ๋ž˜์„œ ๋‹ค๋“ค ์ด ์ž˜ ์•Œ๋ ค์ง„ ๊ธฐ์ˆ  ๊ตฌ์กฐ์— ๊ธฐ๋Œ€์–ด ์—ฌ๊ธฐ์— ๋›ฐ์–ด๋“ค๋ฉฐ ๋‹ค์–‘ํ•œ ์‹คํ—˜์„ ํ•˜๊ณ  ์žˆ๊ฒ ์ง€๋งŒ... ๊ทธ๋งŒํผ PMF๋ฅผ ์ฐพ์ง€ ๋ชปํ•˜๊ณ  ์• ๋งคํ•œ ๊ฒฝ์šฐ ๋˜ํ•œ ๋งŽ์„๊ฑฐ๋ผ ์ถ”์ธกํ•ด ๋ด…๋‹ˆ๋‹ค. ๋น„์Šทํ•œ ์ถœ๋ฐœ์„ ์„ ์ œ๊ณตํ•˜๋Š” ๊ธฐ์ˆ ์ด ๋ฌธ์ œ๊ฐ€ ์•„๋‹ˆ๋ผ ๋‹ค๋ฅธ๊ฒŒ ๋ฌธ์ œ์ผ์ง€, ๊ทธ ํ† ๋Œ€๊ฐ€ ๋˜๋Š” ๊ธฐ์ˆ ์˜ ํ’ˆ์งˆ์ด ๋ฌธ์ œ์ธ๊ฑด์ง€๋Š” ์•„์ง ๋ถˆ๋ถ„๋ช…ํ•ฉ๋‹ˆ๋‹ค.

๊ทธ๋ž˜์„œ ์ด AI Hype์„ ์ด์–ด๊ฐ€๋ ค๋ฉด Gemini์˜ ์—ญํ• ์ด ์ค‘์š”ํ•˜๋‹ค๊ณ  ๋А๋‚๋‹ˆ๋‹ค. ๋‹น์žฅ์€ ์•„๋‹ˆ์ง€๋งŒ ์—ฌ๊ธฐ์— ๋” ๊ธฐ๋ฆ„/๊ธฐ๋Œ€๋ฅผ ๋ถ€์œผ๋ฉฐ ๋‹ค์Œ์„ ์ƒ์ƒํ•  ์ˆ˜ ์žˆ๊ฒŒ ํ•ด์ค˜์•ผ ํ•˜๋‹ˆ๊นŒ์š”.

ํ† ์ด ๋ชจ๋ธ์„ ํ†ตํ•ด์„œ ํ•ด์„๊ฐ€๋Šฅ์„ฑ์„ ๋†’์ด๋Š” ์ผ์—๋„ ๋‹ค์–‘ํ•œ ์‹œ๋„๊ฐ€ ์žˆ์–ด์™”์Šต๋‹ˆ๋‹ค.

Anthropic์˜ Transformer Circuit( https://transformer-circuits.pub/ )์ด๋‚˜

Microsoft research์˜ TinyStories: How Small Can Language Models Be and Still Speak Coherent English? ( https://huggingface.co/papers/2305.07759 ),

๊ทธ๋ฆฌ๊ณ  ์ง€๋‚œ ์ฃผ์˜ Textbooks Are All You Need( https://huggingface.co/papers/2306.11644 ) ๊ฐ™์€ ์†Œ์‹์„ ๋ณด๋ฉด...

์ ์–ด๋„ ๊ณต๊ฐ๋Œ€๋ฅผ ํ˜•์„ฑํ•˜๋Š” ๊ฒƒ์ด '๊ฐ€๋น„์ง€ ์ธ ๊ฐ€๋น„์ง€ ์•„์›ƒ'์˜ ๋ฐ˜๋Œ€๋กœ '์ง€์ ์ธ ๋‚ด์šฉ์ด ์••์ถ•๋˜์–ด์žˆ๋Š” ์–‘์งˆ์˜ ํ† ํฐ์„ ์ž…๋ ฅ์œผ๋กœ ์ฃผ๋Š” ๊ฒƒ์— ๊ด€ํ•œ ์ค‘์š”ํ•จ ๊ฐ™์Šต๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๊ณ ๋ณด๋‹ˆ ์ผ๋ฆฌ์•ผ ์ˆ˜์ธ ์ผ€๋ฒ„๋„ ์ธํ„ฐ๋ทฐ์—์„œ ๊ทธ๋Ÿฐ ์–‘์งˆ์˜ ํ† ํฐ์„ ์–ด๋–ป๊ฒŒ ๊ณ„์† ๋” ๋งŽ์ด ํ™•๋ณดํ•  ๊ฒƒ์ด๋ƒ์— ๊ด€ํ•œ ์ด์•ผ๊ธฐ๋ฅผ ๋ช‡ ๋ฒˆ ํ–ˆ๋˜ ๋“ฏ ํ•ฉ๋‹ˆ๋‹ค.

๊ทธ๋ž˜์„œ ์—ฌ์ „ํžˆ ์ด LLM์˜ ์ •์ฒด๊ฐ€ ๋‹ค ๊ทœ๋ช…๋œ ๊ฒƒ์€ ์•„๋‹ˆ๊ณ  ์—ฐ๊ตฌ๊ฐ€ ๊ณ„์† ๋˜์–ด์•ผ ํ•˜๊ฒ ์ง€๋งŒ... ์ด ๋ฐ”๋‹ฅ์„ ํŒŒ๊ณ ๋“œ๋Š” ํ•ด์„ ๊ฐ€๋Šฅ์„ฑ ์—ฐ๊ตฌ๊ฐ€ ๋‹น์žฅ์˜ ์„ฑ๋Šฅ ํ–ฅ์ƒ์— ํฌ๊ฒŒ ๊ธฐ์—ฌํ•˜์ง„ ์•Š์„ ๊ฒƒ ๊ฐ™์Šต๋‹ˆ๋‹ค. ๋ฌผ๋ก , ๊ฒฝ๋Ÿ‰ํ™” ํ•ด์„œ ์„ฑ๋Šฅ์„ ๋†’์ด๋Š” ๊ฒƒ์—๋Š” ์–ด์ฉŒ๋ฉด ์ข‹์€ ๊ธฐ์—ฌ๋ฅผ ํ•˜๊ฒ ์ฃ .

๊ทธ๋ ‡๋‹ค๊ณ  ๊ต๊ณผ์„œ์— ๋น„์œ ํ•  ์ˆ˜ ์žˆ๋Š” ์–‘์งˆ์˜ ํ† ํฐ๋งŒ ํ•™์Šตํ•œ๋‹ค๊ณ  ๋  ์ผ์ธ๊ฐ€ ํ•˜๋ฉด ์•„๋งˆ ๊ทธ๋ ‡์ง€ ์•Š์„ ๊ฒ๋‹ˆ๋‹ค. '์•ผ์ƒ'์— ๊ทธ๋ฆฌ๊ณ  '๋…ธ์ด์ฆˆ' ์†์— ์žˆ๋Š” ๋ฌด์—‡์ธ๊ฐ€๋„(์‹ฌ์ง€์–ด ๊ทธ๊ฒŒ ํŽธํ–ฅ์ผ์ง€๋ผ๋„) ๋ฌด์ฒ™ ์ค‘์š”ํ•  ์ˆ˜ ์žˆ์œผ๋‹ˆ๊นŒ์š”. ๊ทธ๊ฒŒ '๊ฑฐ๋Œ€'๋ฅผ ํ†ตํ•ด ํš๋“ํ•  ์ˆ˜ ์žˆ์—ˆ๋˜ ๋Šฅ๋ ฅ์˜ ์˜๋ฏธ์ด๊ธฐ๋„ ํ–ˆ๋‹ค๊ณ  ๋ด…๋‹ˆ๋‹ค.

๋˜ํ•œ ์•„์ง LLM๋งŒ์œผ๋กœ๋Š” ํ•ด๊ฒฐํ•  ์ˆ˜ ์—†๋Š” ์›”๋“œ ๋ชจ๋ธ์— ๊ด€ํ•œ ๊ณ„์—ด๋กœ๋Š” ์–€ ๋ฅด์ฟค ๋“ฑ์ด ๊พธ์ค€ํžˆ ์ด์•ผ๊ธฐ ํ•˜๊ณ  ์žˆ๋Š” I-JEPA: The first AI model based on Yann LeCunโ€™s vision for more human-like AI ( https://ai.facebook.com/blog/yann-lecun-ai-model-i-jepa ) ๊ฐ™์€ ๋ฐฉํ–ฅ๋„ ์žˆ๊ณ ,

From Word Models to World Models: Translating from Natural Language to the Probabilistic Language of Thought( https://github.com/gabegrand/world-models )์ด LLM์ด ์ž์—ฐ์–ด ์ž…๋ ฅ์— ๋Œ€ํ•œ ์ถœ๋ ฅ์œผ๋กœ ๋ฒ ์ด์ฆˆ ์ถ”๋ก  ๊ธฐ๋ฐ˜์˜ probabilistic language of thought, ๋˜๋Š” probabilistic programming language ๋ฅผ ์ƒ์„ฑํ•˜๊ฒŒ ํ•˜๋Š” ์ ‘๊ทผ๋„ ๋‚˜์˜ค๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค๋งŒ ์ด๊ฑด ์•ž์— ์–ธ๊ธ‰ํ•œ ๋ฒ ์ŠคํŠธ ํ”„๋ž™ํ‹ฐ์Šค์— ํ•ด๋‹นํ•˜๋Š” ์ด์•ผ๊ธฐ๋กœ๋„ ์—ฌ๊ฒจ์ง‘๋‹ˆ๋‹ค. ( ์œ„ ๋…ผ๋ฌธ์—์„  Lisp์˜ ๋ฐฉ์–ธ์ธ Church๋ฅผ ์‚ฌ์šฉํ–ˆ์ง€๋งŒ, https://probmods.org ๋Š” Javascript ๋น„์Šทํ•œ WebPPL๋กœ ๋„˜์–ด๊ฐ„๋“ฏ ํ•˜๋„ค์š”. ์กฐ์Šˆ์•„ ํ…Œ๋„จ๋ฐ”์›€ ์ชฝ์˜ ์—ฐ๊ตฌ์ž…๋‹ˆ๋‹ค)

๋ฌผ๋ก  ์ด ๋ฒ ์ด์ฆˆ ๊ธฐ๋ฐ˜์œผ๋กœ AI ๋Šฅ๋ ฅ์„ ์ œํ•œํ•˜์ž๋Š” ๊ฒƒ์€ ์š”์Šˆ์•„ ๋ฒค์ง€์˜ค๊ฐ€ ๊ฐ•์กฐํ•˜๋Š” ๋ถ€๋ถ„๊ณผ ๊ณต๋ช…ํ•˜๊ธฐ๋„ ํ•˜๋Š” ์ ์ด ํฅ๋ฏธ๋กญ๊ฒŒ ๋А๊ปด์ง€๊ธด ํ•ฉ๋‹ˆ๋‹ค. AI Scientists: Safe and Useful AI? ( https://yoshuabengio.org/2023/05/07/ai-scientists-safe-and-useful-ai/ )

ํ•˜์ง€๋งŒ ์ด๋Ÿฌํ•œ ์ ‘๊ทผ๋“ค ๋˜ํ•œ ์•„์ง '์™€์šฐ!'๊นŒ์ง„ ์•„๋‹ˆ์ฃ . ๊ฐ€์ƒ ๊ณต๊ฐ„์˜ ์—์ด์ „ํŠธ๋‚˜ ๋กœ๋ด‡์œผ๋กœ ํ™˜๊ฒฝ ์†์—์„œ ์ฒดํ™”๋ฅผ ์‹œ๋„ํ•˜๋Š” ์ ‘๊ทผ๋„ ์žˆ์Šต๋‹ˆ๋‹ค๋งŒ... ์—ญ์‹œ ์ด๊ฒŒ ์‹ค์ƒํ™œ์˜ ์œ ์šฉํ•œ ์„œ๋น„์Šค๋‚˜ ์ œํ’ˆ์œผ๋กœ ๋งŒ๋‚˜๊ธฐ ๊นŒ์ง„ ์‹œ๊ฐ„์ด ๊ฝค ๊ฑธ๋ฆด ๋“ฏ ํ•ฉ๋‹ˆ๋‹ค.

AI ๋ฆฌ์Šคํฌ์— ๊ด€ํ•œ ์–ผ๋ผ์ธ๋จผํŠธ/์•ˆ์ „/๊ทœ์ œ ์ชฝ๋„ ์–ด๋А์ •๋„ ๋…ผ์˜์˜ ๊ฐ€๋‹ฅ์ด ์žกํžˆ๋ฉฐ '์™€์šฐ!'๋Š” ์ค„์–ด๋“ค๊ณ  ์žˆ๋‹ค๊ณ  ์—ฌ๊ฒจ์ง‘๋‹ˆ๋‹ค.
๐Ÿ‘2
์–ด๋–ป๊ฒŒ ๋ณด๋ฉด AI ๋ฆฌ์Šคํฌ๋ฅผ ํ†ตํ•ด์„œ๋„ Hype์„ ์ด์–ด๊ฐ€๊ณ  ์žˆ์—ˆ๋Š”๋ฐ ๋ง์ด์ฃ .

์—ฌ๊ธฐ์— ๊ด€ํ•œ ์ œ ๊ด€์ „ ํฌ์ธํŠธ๋Š”...

The AI Apocalypse: A Scorecard How worried are top AI experts about the threat posed by large language models like GPT-4? ( https://spectrum.ieee.org/artificial-general-intelligence )

Munk Debate on Artificial Intelligence ( https://munkdebates.com/livestreamai )

์˜ ์ธ์„ ์„ ๋ณด๋“ฏ, ๋ช‡ ๊ฐ€์ง€ ์ž˜ ์•Œ๋ ค์ง„ ์ž…์žฅ๋“ค์ด ์žˆ๋Š”๋ฐ... ๊ตณ์ด ์ด ์ž…์žฅ๋“ค์„ ํ˜‘์˜ํ•ด์„œ ํ•˜๋‚˜๋กœ ํ•  ํ•„์š”์—†์ด,

OpenAI ๋“ฑ์—์„œ ์ด์•ผ๊ธฐํ•˜๋Š” ์–ผ๋ผ์ธ๋จผํŠธ์— ๋Œ€ํ•œ ์ด์Šˆ๋Š” ๊ทธ์ชฝ ๊ทธ๋ฃน์—์„œ,
์ œํ”„๋ฆฌ ํžŒํŠผ ๋“ฑ์ด ๊ฑฑ์ •ํ•˜๋Š” ๋ฐ”์— ๋Œ€ํ•œ ๋Œ€์•ˆ์€ ๊ทธ์ชฝ ๊ทธ๋ฃน์—์„œ,
์š”์Šˆ์•„ ๋ฒค์ง€์˜ค ๋“ฑ์ด ๊ฑฑ์ •ํ•˜๋Š” ๋ฐ”์— ๋Œ€ํ•œ ๋Œ€์•ˆ์€ ๊ทธ์ชฝ ๊ทธ๋ฃน์—์„œ,
์žฅ๊ธฐ์ฃผ์˜์ž/ํšจ์œจ์  ์ดํƒ€์ฃผ์˜์ž ๋“ฑ์˜ ๊ณ„์—ด์ด ๊ฑฑ์ •ํ•˜๋Š” ๋ฐ”์— ๋Œ€ํ•œ ๋Œ€์•ˆ์€ ๊ทธ์ชฝ ๊ทธ๋ฃน์—์„œ,
๋ฉœ๋ผ๋‹ˆ ๋ฏธ์ฒผ, ํ™•๋ฅ ์  ์•ต๋ฌด์ƒˆ ์ €์ž ๊ณ„์—ด์ด ๊ฑฑ์ •ํ•˜๋Š” ๋ฐ”๋Š” ๊ทธ์ชฝ ๊ทธ๋ฃน์—์„œ,
์—˜๋ฆฌ์—์ € ์œณ์ฝ”์šฐ์Šคํ‚ค๋‚˜ ๊ฒŒ๋ฆฌ ๋งˆ์ปค์Šค ๊ณ„์—ด์€ ๋˜ ๊ทธ์ชฝ ๋‚˜๋ฆ„์œผ๋กœ,
์–€ ๋ฅด์ฟค ๋“ฑ์˜ ๊ทธ๋ฃน๋„ ์—ญ์‹œ ๊ทธ์ชฝ ๋‚˜๋ฆ„์˜ ํ™”๋‘๋กœ,
์•ค๋“œ๋ฅ˜ ์‘์ด ๋งํ•˜๋Š” ๊ฒƒ ์ฒ˜๋Ÿผ ๊ณต๊ฐ๋Œ€๋ฅผ ํ˜•์„ฑํ•˜๋Š” ๊ฒƒ์€ ๋˜ ์•ค๋“œ๋ฅ˜ ์‘ ๋“ฑ์˜ ๊ทธ๋ฃน์—์„œ,
EU๋‚˜ ๋ฏธ์˜ํšŒ ์ชฝ์˜ ๊ทœ์ œ ๋…ผ์˜๋„ ๋‚˜๋ฆ„์˜ ๋ฐฉ์‹์œผ๋กœ~

๊ฐ์ž์˜ ์ž…์žฅ์—์„œ ๊ฐ์ž ์šฐ๋ คํ•˜๋Š” '๋ฆฌ์Šคํฌ'๋ฅผ ์™„ํ™”ํ•˜๋Š” ๋„์ „์„ ํ•˜๊ณ  ๊ทธ ์„ฑ๊ณผ๋ฅผ ๊ณต์œ ํ•˜๋ฉด Anthropic์ด ์ด์•ผ๊ธฐํ•œ ๋‹ค์–‘ํ•œ ์ˆ˜์ค€์˜ 'AI ๋ฆฌ์Šคํฌ'์— ๊ฐ๊ฐ ๋Œ€์‘ํ•˜๋Š” ํ—ท์ง•์„ ํ•˜๋Š” ์ ‘๊ทผ, Core Views on AI Safety: When, Why, What, and How( https://www.anthropic.com/index/core-views-on-ai-safety )๊ณผ ๋น„์Šทํ•ด์ง€๋Š” ๋ชจ์–‘์ƒˆ๋ผ...

์ด์ œ ๋‚˜์˜ฌ๋ฒ•ํ•œ ๋…ผ์˜๋Š” ๊ฝค ๋‚˜์™”๊ณ  ์ด ๋˜ํ•œ ๋‹น๋ถ„๊ฐ„์€ '์™€์šฐ!'๋Š” ์—†์„ ๊ฒƒ ๊ฐ™๋‹จ ์ „๋ง์„ ํ•˜๊ฒŒ๋œ๋‹ฌ๊นŒ์š”. ์ •๋ง AI ๋ฆฌ์Šคํฌ๊ฐ€ ํฌ๊ฒŒ ํ„ฐ์ง€๋Š” ์‚ฌ๊ฑด์ด ์ผ์–ด๋‚œ๋‹ค๋ฉด ์ด AI ๋ฆฌ์Šคํฌ ์ด์Šˆ๊ฐ€ ํฌ๊ฒŒ ๋ถ€๊ฐ๋˜๋ฉฐ AI Hype์— ์ผ์กฐํ•˜๊ฒ ์ง€๋งŒ ๊ทธ๋Ÿฌ์ง€ ์•Š๋Š”๋‹ค๋ฉด, ์•„์ด๋Ÿฌ๋‹ˆํ•˜๊ฒŒ๋„ ์ด ๋˜ํ•œ ์ •์ฒด๊ฐ์— ์ผ์กฐํ•  ์ˆ˜ ์žˆ๊ฒ ์Šต๋‹ˆ๋‹ค.

๊ธฐ๋ฐ˜์ด ๋˜๋Š” ๊ธฐ์ˆ  ์ง„๋ณด์—์„œ '์™€์šฐ!'๊ฐ€ ์—†๋‹ค๋ฉด ์—ฌ๊ธฐ์— ํ† ๋Œ€๋ฅผ ๋‘๋Š” ์˜ˆ์ˆ /ํผํฌ๋จผ์Šค ๋ถ„์•ผ์—์„œ๋„ ์•„์ง ์ด๊ฑธ ์ ‘๋ชฉํ•ด ๋ณด์ง€ ์•Š์€ ๋„๋ฉ”์ธ์ด๋‚˜ ์žฅ๋ฅด์™€ ๋งŒ๋‚จ์ด ์žˆ์„ ๋•Œ ์ž ๊น ์ž ๊น ์ด์Šˆ๊ฐ€ ๋งŒ๋“ค์–ด์ง€๊ฒ ์ง€๋งŒ, ์–ด๋””์„œ ๋ดค๋˜ ๊ธฐ์‹œ๊ฐ์ด ๋“ค๋ฉฐ ์ •์ฒด๊ฐ์ด ์ผ์–ด๋‚  ์ˆ˜ ์žˆ์„ ๊ฒ๋‹ˆ๋‹ค.

๋งˆ์ฐฌ๊ฐ€์ง€๋กœ ์—ญ์‹œ ํ† ๋Œ€๊ฐ€ ๋˜๋Š” ๊ธฐ์ˆ  ์ง„๋ณด์—์„œ '์™€์šฐ'๊ฐ€ ์—†๋‹ค๋ฉด ์—ฌ๊ธฐ์— ํ† ๋Œ€๋ฅผ ๋‘๋Š” '์ฑ…'๋„ ๋น ๋ฅด๊ฒŒ ๊ด€์‹ฌ์ด ๋ถˆํƒ€์˜ค๋ฅธ ๊ฒƒ ์ฒ˜๋Ÿผ, ๋ฒ ์ŠคํŠธ ์…€๋Ÿฌ ์–‘์„œ ๋ช‡ ๊ถŒ์ด ์ •์ ์„ ํ•œ๋ฒˆ ์ฐ์œผ๋ฉฐ... ๋งŽ์€ ์‚ฌ๋žŒ๋“ค์ด ์–ด๋А ์ •๋„ ๊ฐ์„ ์žก์€ ๋’ค์—๋Š” ๋น ๋ฅด๊ฒŒ ํƒ€์˜ค๋ฅธ ๊ฒƒ ๋งŒํผ ๋น ๋ฅด๊ฒŒ ์‹์„ ์ˆ˜๋„ ์žˆ๊ณ ์š”. ๋ฌผ๋ก  ๋งŽ์€ ์‚ฌ๋žŒ๋“ค์ด ๊ฐ์„ ์žก๊ธฐ๊นŒ์ง€๋Š” ๋กฑํ…Œ์ผ์ด ์žˆ๊ฒ ์ฃ .

๋”ฐ๋ผ์„œ... ๋‹ค์‹œ ํ•œ๋ฒˆ '์™€์šฐ!' ๋ชจ๋จผํŠธ, ์ด ์ •๋„๊ฐ€ ๋˜๋Š” ๊ฑด ์ฒ˜์Œ์ด์•ผ!๋ผ๋Š” '๊ฒฝ์ด๊ฐ'์„ ์ฃผ๋Š” ๊ฒƒ์ด ๋‹ค์‹œ ์ ๋‹นํ•œ ํƒ€์ด๋ฐ์— ์ฝค๋ณด๋ฅผ ์ด์–ด๊ฐ€ ์ฃผ์ง€ ์•Š๋Š”๋‹ค๋ฉด, '์ •์ฒด๊ฐ'์„ ๋А๋ผ๋ฉฐ, ์ผ์ƒ์œผ๋กœ๋“ค ๋Œ์•„๊ฐ€๊ธฐ ์‹œ์ž‘ํ•  ์ˆ˜ ์žˆ๊ฒ ์ฃ .

๋ฌผ๋ก  ๊ทธ ์‚ฌ์ด ์ด๋ฏธ ๋ฌด์—‡์ธ๊ฐ€๊ฐ€ ๋‰ด๋…ธ๋ง์ด ๋˜์—ˆ์„ ๊ฒ๋‹ˆ๋‹ค. ๋งค์šฐ ์ค‘์š”ํ•œ ๊ฑด๋ฐ, ์ด์   ๊ทธ๋ ‡๊ฒŒ ์‹ ๊ธฐํ•˜์ง€ ์•Š๊ณ  ๋„ˆ๋ฌด ๋‹น์—ฐํ•œ ๊ทธ๋Ÿฐ ๋ฌด์—‡์ธ๊ฐ€๋กœ์š”.

์–ด์ฉŒ๋ฉด... ๊ทธ๋ƒฅ ์ œ๊ฐ€ AI ๋‰ด์Šค์— ๊ด€ํ•ด ํ”ผ๋กœ๊ฐ์„ ๋งŽ์ด ๋А๋ผ๊ณ  ์žˆ๋Š” ์ค‘์ด๊ฑฐ๋‚˜... ๋…ธํšŒํ•ด์„œ ์†Œ์‹์— ๋ฏผ๊ฐํ•˜๊ฒŒ ๊ฐ์‘ํ•˜๋Š” ๋Šฅ๋ ฅ์ด ๋–จ์–ด์ ธ์žˆ๋Š” ์ƒํƒœ์ผ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๋‰ด์Šค๋Š” ๊ณ„์† ๊ฐ๋‹นํ•˜๊ธฐ ์–ด๋ ต๊ฒŒ ๋งŽ์ด ๋‚˜์˜ฌ๊ฒ๋‹ˆ๋‹ค. ๋‹ค๋งŒ ๊ฑฐ๊ธฐ์— ๊ด€ํ•œ ๋ฐ˜์‘์ด ์ •์ฒด๋˜์–ด ์žˆ๋Š” ๊ฒƒ์ด์ฃ .

๊ทธ๋ž˜๋„ ์ด๋ ‡๊ฒŒ ์ •์ฒด๊ฐ์„ ๋А๋ผ๋Š”๊ฒŒ ๊ณ„์† ๊ณ„์† ๋‹ค์Œ์˜ ํฐ ์ž๊ทน์„ ์ฐพ๋Š” ๋„ํŒŒ๋ฏผ ์ค‘๋…์ด ํ•ด๋…๋˜๊ณ  ์žˆ๋Š” ์ค‘์ด๋ผ๋ฉด... ๋‹คํ–‰์ž…๋‹ˆ๋‹ค.
๐Ÿ‘1