Continuous Learning_Startup & Investment
2.43K subscribers
513 photos
5 videos
16 files
2.76K 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
Wanting to get started with Generative AI and LLMs, but not sure where to start? ๐Ÿค” I am super excited to share Amazon Web Services (AWS) and DeepLearning.AI just launched "Generative AI with LLMs" course, designed specifically for individuals and beginners! ๐Ÿ”ฐ๐Ÿ”ฅ

In Generative AI with Large Language Models (LLMs), youโ€™ll learn the fundamentals of how generative AI works and how to use the Hugging Face ecosystem (Transformers, PEFT, TRL) to instruction-tune, RLHF, or deploy open-source LLMs! ๐Ÿคฏ

๐Ÿ‘‰ https://lnkd.in/ep68k-Pk

I am incredibly proud to say that I worked behind the scenes with Antje Barth, Chris Fregly, and Mike Chambers to make this course a reality. Huge kudos to everyone who was involved.
๐Ÿค— https://lnkd.in/e3a8jXw7

If you've ever been curious about how generative AI works or want to refresh your knowledge, this course is an absolute must-attend! ๐Ÿ”ฅ๐Ÿค
Forwarded from ์š”์ฆ˜AI
์ตœ์ดˆ๋กœ IOS ํƒˆ์˜ฅ์— ์„ฑ๊ณตํ–ˆ๋˜ ๋ฏธ๊ตญ์˜ ์ฒœ์žฌ ํ•ด์ปค ์กฐ์ง€ ํ˜ธ์ธ (George Hotz)๊ฐ€ ๊ทธ๋™์•ˆ ๋ฒ ์ผ์— ๊ฐ์ถฐ์ง„ GPT-4์˜ ๊ตฌ์กฐ์— ๋Œ€ํ•ด ์–ธ๊ธ‰ํ–ˆ์Šต๋‹ˆ๋‹ค.

๊ทธ๊ฐ€ GPT-4์˜ ์„ฑ๋Šฅ์„ ๋†’์ด๋Š” ํ•ต์‹ฌ ๊ตฌ์กฐ๋กœ ์–ธ๊ธ‰ํ•œ โ€˜MoE(Mixture of Experts)โ€™ ๋ชจ๋ธ์— ๋Œ€ํ•ด ์•Œ๊ธฐ ์‰ฝ๊ฒŒ ๋‚ด์šฉ์„ ์ •๋ฆฌํ•ด๋ณด์•˜์Šต๋‹ˆ๋‹ค.

https://news.aikoreacommunity.com/ceonjaehaekeo-jiohasi-gpt-4yi-bimileul-puleonaeda/

1/ ์กฐ์ง€ ํ˜ธ์ธ ๋Š” OpenAI์˜ GPT-4๊ฐ€ 1์กฐ ๊ฐœ์˜ ํŒŒ๋ผ๋ฏธํ„ฐ(parameter)๋ฅผ ๊ฐ€์ง„ ๋ชจ๋ธ์ด ์•„๋‹Œ, 2,200์–ต ๊ฐœ์˜ ํŒŒ๋ผ๋ฏธํ„ฐ๋ฅผ ๊ฐ€์ง„ ๋ชจ๋ธ 8๊ฐœ๊ฐ€ ํ˜ผํ•ฉ๋œ ๊ตฌ์กฐ๋กœ ์ด๋ฃจ์–ด์ ธ ์žˆ๋‹ค๊ณ  ์ฃผ์žฅ.

์ฆ‰ ๊ฐ™์€ ํฌ๊ธฐ์˜ ๋ชจ๋ธ์„ ์—ฌ๋Ÿ ๋ฒˆ ํ›ˆ๋ จ ์‹œํ‚จ ํ›„, โ€˜MoEโ€™๋ผ๋Š” ๋ชจ๋ธ ๊ตฌ์กฐ๋ฅผ ํ™œ์šฉํ•˜์—ฌ 1์กฐ ๊ฐœ์˜ ํŒŒ๋ผ๋ฏธํ„ฐ๋ฅผ ๊ฐ€์ง„ ํฐ ๋ชจ๋ธ์ธ ์ฒ™ ํŠธ๋ฆญ์„ ์ผ๋‹ค๋Š” ๊ฒƒ.

๊ทธ๋ ‡๋‹ค๋ฉด MoE๊ฐ€ ๋ฌด์—‡์ผ๊นŒ?

2/ MoE(Mixture of Experts)๋Š” ์—ฌ๋Ÿฌ ๊ฐœ์˜ ์‹ ๊ฒฝ๋ง์„ ์„œ๋กœ ๋‹ค๋ฅธ ๋ถ„์•ผ์— ํŠนํ™”๋œ ์ „๋ฌธ๊ฐ€(Experts) ์‹ ๊ฒฝ๋ง์œผ๋กœ ๊ฐ๊ฐ ํ›ˆ๋ จ์‹œํ‚ค๊ณ , ์ด ์‹ ๊ฒฝ๋ง๋“ค์„ ํ˜ผํ•ฉ(Mixture)ํ•˜์—ฌ ํ™œ์šฉํ•˜๋Š” ๋”ฅ๋Ÿฌ๋‹ ๋ชจ๋ธ ๊ตฌ์กฐ์ž„.

์ฆ‰ ์—ฌ๋Ÿฌ ๊ฐœ์˜ ์„œ๋กœ ๋‹ค๋ฅธ ์‹ ๊ฒฝ๋ง(์ „๋ฌธ๊ฐ€)์ด ์„œ๋กœ ๋‹ค๋ฅธ ๋ฌธ์ œ๋‚˜ ๋ฐ์ดํ„ฐ ๋ถ„์•ผ๋ฅผ ์ฒ˜๋ฆฌํ•˜๋„๋ก ์„ค๊ณ„๋œ ๋ชจ๋ธ์ธ ๊ฒƒ.

3/ MoE ๋ชจ๋ธ์€ ํฌ๊ฒŒ ๋‘ ๊ฐ€์ง€๋กœ ๊ตฌ์„ฑ๋จ. ์ „๋ฌธ๊ฐ€(Experts)์™€ ๊ฒŒ์ดํŠธ(Gate).

์ „๋ฌธ๊ฐ€๋Š” ์•ž์„œ ๋งํ–ˆ๋“ฏ ํŠนํ™”๋œ ๊ฐ ๋ถ€๋ถ„์— ๋Œ€ํ•œ ์ฒ˜๋ฆฌ๋ฅผ ๋‹ด๋‹นํ•จ. ๊ฒŒ์ดํŠธ๋Š” ์ž…๋ ฅ๊ฐ’(input)์— ๋Œ€ํ•ด ๊ฐ ์ „๋ฌธ๊ฐ€์—๊ฒŒ ๊ฐ€์ค‘์น˜๋ฅผ ๋ถ€์—ฌํ•˜๋Š” ์—ญํ• ์„ ์ˆ˜ํ–‰ํ•จ.

4/ MoE ๋ชจ๋ธ์ด ๋‹ต๋ณ€์„ ๋‚ด๋Š” ๋ฐฉ์‹์€ ๋‹ค์–‘ํ•จ. ํฐ ๊ฐ€์ค‘์น˜๋ฅผ ๋ถ€์—ฌ๋ฐ›์€ ์ „๋ฌธ๊ฐ€๊ฐ€ ์ถœ๋ ฅ๊ฐ’(output)์„ ์ƒ์„ฑํ•˜๊ฑฐ๋‚˜, ํ˜น์€ ๊ฐ ์ „๋ฌธ๊ฐ€๋“ค์˜ ๋‹ต๋ณ€์— ๊ฐ€์ค‘์น˜๋ฅผ ๋งค๊ธด ํ›„ ์ด๋ฅผ ํ•ฉ์ณ์„œ ์ถœ๋ ฅ๊ฐ’์„ ์ƒ์„ฑํ•˜๋Š” ๋ฐฉ์‹๋„ ์žˆ์Œ.

์–ด๋–ค ๋ฐฉ์‹์ด๋“  ๊ฐ ๋ชจ๋ธ์ด ์ „๋ฌธํ™”๋œ ๋ถ„์•ผ์— ํŠนํ™”๋œ ๋‹ต๋ณ€์„ ์ œ๊ณตํ•˜๊ธฐ ๋•Œ๋ฌธ์—, ๊ฐ™์€ ํฌ๊ธฐ์˜ ๋ชจ๋ธ๋กœ ์ด๋ฃจ์–ด์ง„ ๋‹จ์ผ ์‹ ๊ฒฝ๋ง ๊ตฌ์กฐ๋ณด๋‹ค ๋” ๋†’์€ ๋‹ต๋ณ€ ์„ฑ๋Šฅ์„ ๋‚ผ ์ˆ˜ ์žˆ๋Š” ๊ฒƒ.

5/ ํ•œ ๊ฐ€์ง€ ๋‹จ์ ์€ ์—ฌ๋Ÿฌ ๊ฐœ์˜ ๋ชจ๋ธ์„ ํ•œ ๋ฒˆ์— ์‚ฌ์šฉํ•˜๋‹ค ๋ณด๋‹ˆ ๊ณ„์‚ฐ ๋ฐ ๋ฉ”๋ชจ๋ฆฌ ๋น„์šฉ์ด ๊ธฐ์กด ๋‹จ์ผ ์‹ ๊ฒฝ๋ง ๊ตฌ์กฐ๋ณด๋‹ค ๋†’์•„์งˆ ์ˆ˜ ์žˆ๋‹ค๋Š” ๊ฒƒ.(MoE๋Š” ์—ฌ๋Ÿฌ ๊ฐœ์˜ ์‹ ๊ฒฝ๋ง ๋ชจ๋ธ์„ ์ด์šฉํ•˜๊ธฐ ๋•Œ๋ฌธ)

ํ•˜์ง€๋งŒ ์ด๋Š” ๋ณ‘๋ ฌ ์ฒ˜๋ฆฌ์™€ ๊ด€๋ จ๋œ ์—ฐ๊ตฌ์™€, Sparse Gate ๋“ฑ ๋‹ค์–‘ํ•œ ์—ฐ๊ตฌ๋ฅผ ํ†ตํ•ด ๊ณ„์†ํ•ด์„œ ํ•ด๊ฒฐ๋˜๊ณ  ์žˆ์Œ. ์•„๋ž˜ ๋งํฌ๋Š” ๊ด€๋ จ ์—ฐ๊ตฌ ๋…ผ๋ฌธ ์ค‘ ํ•˜๋‚˜.
https://arxiv.org/pdf/2212.05055.pdf

6/ ๋˜ํ•œ MoE ๊ตฌ์กฐ๋Š” ํŒŒ๋ผ๋ฏธํ„ฐ๊ฐ€ ์ ์€ ๋ชจ๋ธ์ผ์ˆ˜๋ก ํ•œ ๊ฐœ์˜ ์‹ ๊ฒฝ๋ง์œผ๋กœ ์ž‘๋™ํ•˜๋Š” ๋ชจ๋ธ๋ณด๋‹ค ์„ฑ๋Šฅ์„ ํ–ฅ์ƒ์‹œํ‚ค๊ธฐ ์–ด๋ ค์›€.

๋ฐ์ดํ„ฐ ์„ธํŠธ๊ฐ€ ์ž‘์„์ˆ˜๋ก ๊ฐ ๋ชจ๋ธ์ด ํ•™์Šตํ•  ์ˆ˜ ์žˆ๋Š” ํŠน์ • ๋ฐ์ดํ„ฐ๊ฐ€ ๋งค์šฐ ์ œํ•œ์ ์ด๊ธฐ ๋•Œ๋ฌธ.

ํ•˜์ง€๋งŒ ํ•™์Šต๋œ ๋ฐ์ดํ„ฐ ์„ธํŠธ๊ฐ€ ์ปค์งˆ์ˆ˜๋ก ๊ฐ ์‹ ๊ฒฝ๋ง์ด ํ•™์Šตํ•  ์ˆ˜ ์žˆ๋Š” ๋ฐ์ดํ„ฐ๊ฐ€ ๋งŽ์•„์ง€๋ฏ€๋กœ ์ด ๊ตฌ์กฐ๊ฐ€ ๋งค์šฐ ํšจ์œจ์ ์œผ๋กœ ์ž‘์šฉํ•จ.

7/ ์ฆ‰ MoE ๊ตฌ์กฐ๋Š” ๋Œ€๊ทœ๋ชจ ์–ธ์–ด ๋ชจ๋ธ์— ๋งค์šฐ ์ ํ•ฉํ•˜๋‹ค๋Š” ๊ฒƒ.

GPT-3.5์˜ ํŒŒ๋ผ๋ฏธํ„ฐ(parameter)๋Š” 1,750์–ต ๊ฐœ์ด๋ฉฐ, ์กฐ์ง€ ํ˜ธ์ธ ๊ฐ€ GPT-4์—์„œ ์‚ฌ์šฉ๋˜์—ˆ๋‹ค๊ณ  ์ฃผ์žฅํ•œ ๋ชจ๋ธ์˜ ํŒŒ๋ผ๋ฏธํ„ฐ๋Š” 2,200์–ต ๊ฐœ.

๋งŒ์•ฝ ๊ทธ์˜ ์ฃผ์žฅ์ด ๋งž๋‹ค๋ฉด GPT-3.5์™€ GPT-4์˜ ์„ฑ๋Šฅ ์ฐจ์ด๋Š” ํŒŒ๋ผ๋ฏธํ„ฐ์˜ ์ฐจ์ด๊ฐ€ ์•„๋‹Œ MoE ๊ตฌ์กฐ์˜ ์œ ๋ฌด์—์„œ ์˜ค๋Š” ์ฐจ์ด๋ผ๋Š” ๋œป.

8/ MoE๊ฐ€ ๊ฐ–๋Š” ํŠน์„ฑ์€ GPT-4์™€ ๊ฐ™์€ ๋Œ€๊ทœ๋ชจ AI ๋ชจ๋ธ์„ ๋งŒ๋“œ๋Š” ๋ฐ ์ ํ•ฉํ•˜๊ฒŒ ์‚ฌ์šฉ๋  ์ˆ˜ ์žˆ์Œ.

์ƒ˜ ์•ŒํŠธ๋งŒ ๋˜ํ•œ AI ๋ชจ๋ธ์˜ ๊ทœ๋ชจ์˜ ํ•œ๊ณ„์— ๋Œ€ํ•ด์„œ ์–ธ๊ธ‰ํ•œ ๋ฐ”๊ฐ€ ์žˆ๊ธฐ์—, ๊ทœ๋ชจ์˜ ํ•œ๊ณ„๋ฅผ ๋›ฐ์–ด๋„˜๋Š” ์„ฑ๋Šฅ์„ ๋ฐœํœ˜ํ•˜๋„๋ก ํ•˜๋Š” MoE ๊ตฌ์กฐ์˜ ํ™œ์šฉ์„ฑ์ด ๋งค์šฐ ๊ธฐ๋Œ€๋จ.

ํŒŸ์บ์ŠคํŠธ ์ „๋ฌธ์€ ์ด๊ณณ์—์„œ ๋ณผ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์ฝ์–ด์ฃผ์…”์„œ ๊ฐ์‚ฌํ•ฉ๋‹ˆ๋‹ค. :)

https://www.latent.space/p/geohot#details
์ธํ„ฐ๋ทฐ ๋ณด๋Š” ๋‚ด๋‚ด ์ฐฝ์—…์ž๊ฐ€ ์ž์‹ ์ด ๋งŒ๋“ค๊ณ  ์žˆ๋Š” ์ œํ’ˆ๊ณผ ๊ณ ๊ฐ์„ ์‚ฌ๋ž‘ํ•˜๋Š” ๊ฒŒ ๋А๊ปด์ง„๋‹ค._Character.ai

1. ์‚ฌ์šฉ์ž์—๊ฒŒ Character.ai๋ฅผ ์–ด๋–ป๊ฒŒ ์‚ฌ์šฉํ•˜๋ผ๊ณ  ์•Œ๋ ค๋“œ๋ฆฌ๋Š” ๊ฒƒ์€ ์ €ํฌ์˜ ์ผ์ด ์•„๋‹™๋‹ˆ๋‹ค. ์ €ํฌ์˜ ์ผ์€ ์ผ๋ฐ˜์ ์ธ ๊ฒƒ์„ ๋‚ด๋†“๊ณ  ์‚ฌ๋žŒ๋“ค์ด ๊ทธ๊ฒƒ์„ ์ฆ๊ฒ๊ฒŒ ์‚ฌ์šฉํ•˜๋Š” ๊ฒƒ์„ ๋ณด๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค.
2. ๋งŽ์€ ์‚ฌ๋žŒ๋“ค์ด ์™ธ๋กญ๊ฑฐ๋‚˜ ๊ณ ๋ฏผ์ด ์žˆ์–ด ๋Œ€ํ™” ์ƒ๋Œ€๊ฐ€ ํ•„์š”ํ•˜๊ธฐ ๋•Œ๋ฌธ์— ํŽ˜๋ฅด์†Œ๋‚˜๋ฅผ ์‚ฌ์šฉํ•ฉ๋‹ˆ๋‹ค.
3. ์œ ์ €๋“ค์ด Character.ai์˜ ์บ๋ฆญํ„ฐ๋“ค์„ ๋กคํ”Œ๋ ˆ์ž‰ ๊ฒŒ์ž„, ํ…์ŠคํŠธ ์–ด๋“œ๋ฒค์ฒ˜, TV ๋˜๋Š” ์ธํ„ฐ๋„ท ์ธํ”Œ๋ฃจ์–ธ์„œ ์‹œ์ฒญ ๋“ฑ ๋‹ค์–‘ํ•œ ๋ฐฉ์‹์œผ๋กœ ์‚ฌ์šฉํ•ฉ๋‹ˆ๋‹ค.

https://youtu.be/GavsSMyK36w
https://youtu.be/emCoG-hA7AE

Itโ€™s not our job to tell you what uses for. Our job is to put out something general and see people enjoy using it.
Many use personas because they are lonely or troubled and need someone to talk to.
Noam Shazeer talks about the concept of a persona, which is a character or a person that users create in order to use their imagination. He explains that people use persona in various ways, such as role-playing games, text adventures, and watching TV or internet influencers.
the backstory of Character, where they wanted to create a technology that was accessible, flexible and put the user in control.
New open-source LLMs! ๐Ÿ”” Salesforce just released XGen 7B, a new LLM with an 8k context under the Apache 2.0 license. ๐Ÿ”“ XGen uses the same architecture as Metas LLaMa and is, therefore, a 1-to-1 replacement for commercial use! ๐Ÿ”ฅ XGen achieves similar performance to LLaMa on MMLU and outperforms on coding! ๐ŸŽ–

TL;DR; โœจ:
๐Ÿ”  Trained on 1.5T Tokens
๐ŸชŸ 8192 context window
๐Ÿงฎ 7B parameter
๐Ÿ”“ Apache 2.0 license
๐Ÿง  Trained on TPUs
๐Ÿง‘๐Ÿปโ€๐Ÿ’ป Can write code
๐Ÿค— Available on Hugging Face

Model: https://lnkd.in/emHEPZy8
Announcement Blog: https://lnkd.in/e6utBth9

It's exciting to see more LLaMa models released with permissive licenses. Hopefully, Salesforce will continue the model family with 13 or 16B versions.๐Ÿš€
We develop a method to test global opinions represented in language models. We find the opinions represented by the models are most similar to those of the participants in USA, Canada, and some European countries. We also show the responses are steerable in separate experiments.

https://twitter.com/AnthropicAI/status/1674461614056292353?s=20
Inflection AI today announced that the company has raised $1.3 billion in a fresh round of funding led by Microsoft, Reid Hoffman, Bill Gates, Eric Schmidt, and new investor NVIDIA. The new funding brings the total raised by the company to $1.525 billion.

Largest AI cluster in the world
The deployment of 22,000 NVIDIA H100 GPUs in one cluster is truly unprecedented, and will support training and deployment of a new generation of large-scale AI models. Combined, the cluster develops a staggering 22 exaFLOPS in the 16-bit precision mode, and even more if lower precision is utilized. We estimate that if we entered our cluster in the recent TOP500 list of supercomputers, it would be the 2nd and close to the top entry, despite being optimized for AI โ€“ rather than scientific โ€“ applications. The rollout of the cluster is actively under way, and we have already been able to confirm its performance in the recent MLPerf benchmark.

https://inflection.ai/inflection-ai-announces-1-3-billion-of-funding
Consider the future of this decidedly "semantic" AI https://learn.microsoft.com/en-us/semantic-kernel/when-to-use-ai/schillace-laws
The "Schillace Laws" were formulated after working with a variety of Large Language Model (LLM) AI systems to date. Knowing them will accelerate your journey into this exciting space of reimagining the future of software engineering. Welcome!

Donโ€™t write code if the model can do it; the model will get better, but the code won't. The overall goal of the system is to build very high leverage programs using the LLM's capacity to plan and understand intent. It's very easy to slide back into a more imperative mode of thinking and write code for aspects of a program. Resist this temptation โ€“ to the degree that you can get the model to do something reliably now, it will be that much better and more robust as the model develops.

Trade leverage for precision; use interaction to mitigate. Related to the above, the right mindset when coding with an LLM is not "let's see what we can get the dancing bear to do," it's to get as much leverage from the system as possible. For example, it's possible to build very general patterns, like "build a report from a database" or "teach a year of a subject" that can be parameterized with plain text prompts to produce enormously valuable and differentiated results easily.

Code is for syntax and process; models are for semantics and intent. There are lots of different ways to say this, but fundamentally, the models are stronger when they are being asked to reason about meaning and goals, and weaker when they are being asked to perform specific calculations and processes. For example, it's easy for advanced models to write code to solve a sudoku generally, but hard for them to solve a sudoku themselves. Each kind of code has different strengths and it's important to use the right kind of code for the right kind of problem. The boundaries between syntax and semantics are the hard parts of these programs.

The system will be as brittle as its most brittle part. This goes for either kind of code. Because we are striving for flexibility and high leverage, itโ€™s important to not hard code anything unnecessarily. Put as much reasoning and flexibility into the prompts and use imperative code minimally to enable the LLM.

Ask Smart to Get Smart. Emerging LLM AI models are incredibly capable and "well educated" but they lacks context and initiative. If you ask them a simple or open-ended question, you will get a simple or generic answer back. If you want more detail and refinement, the question has to be more intelligent. This is an echo of "Garbage in, Garbage out" for the AI age.

Uncertainty is an exception throw. Because we are trading precision for leverage, we need to lean on interaction with the user when the model is uncertain about intent. Thus, when we have a nested set of prompts in a program, and one of them is uncertain in its result ("One possible way...") the correct thing to do is the equivalent of an "exception throw" - propagate that uncertainty up the stack until a level that can either clarify or interact with the user.

Text is the universal wire protocol. Since the LLMs are adept at parsing natural language and intent as well as semantics, text is a natural format for passing instructions between prompts, modules and LLM based services. Natural language is less precise for some uses, and it is possible to use structured language like XML sparingly, but generally speaking, passing natural language between prompts works very well, and is less fragile than more structured language for most uses. Over time, as these model-based programs proliferate, this is a natural "future proofing" that will make disparate prompts able to understand each other, the same way humans do.
Hard for you is hard for the model. One common pattern when giving the model a challenging task is that it needs to "reason out loud." This is fun to watch and very interesting, but it's problematic when using a prompt as part of a program, where all that is needed is the result of the reasoning. However, using a "meta" prompt that is given the question and the verbose answer and asked to extract just the answer works quite well. This is a cognitive task that would be easier for a person (it's easy to imagine being able to give someone the general task of "read this and pull out whatever the answer is" and have that work across many domains where the user had no expertise, just because natural language is so powerful). So, when writing programs, remember that something that would be hard for a person is likely to be hard for the model, and breaking patterns down into easier steps often gives a more stable result.

Beware "pareidolia of consciousness"; the model can be used against itself." It is very easy to imagine a "mind" inside an LLM. But there are meaningful differences between human thinking and the model. An important one that can be exploited is that the models currently don't remember interactions from one minute to the next. So, while we would never ask a human to look for bugs or malicious code in something they had just personally written, we can do that for the model. It might make the same kind of mistake in both places, but it's not capable of "lying" to us because it doesn't know where the code came from to begin with. _This means we can "use the model against itself" in some places โ€“ it can be used as a safety monitor for code, a component of the testing strategy, a content filter on generated content, etc. _
State of GPT talk by Andrej Karpathy: https://www.youtube.com/watch?v=bZQun8Y4L2A&t=373s

Would highly recommend watching the above! A 45-minute lecture going over the State of Generative LLMs, how are they trained, what they can and can't do, advanced techniques like CoT, ReAct, Reflection, BabyAGI, and Agents in general and finally some great tips on using LLMs in production. Pretty simple but very very informative
Continuous Learning_Startup & Investment
State of GPT talk by Andrej Karpathy: https://www.youtube.com/watch?v=bZQun8Y4L2A&t=373s Would highly recommend watching the above! A 45-minute lecture going over the State of Generative LLMs, how are they trained, what they can and can't do, advanced techniquesโ€ฆ
Here's an https://assembly.ai transcript and chapter summaries:
๐Ÿ‘‚๐Ÿผ ๐Ÿค– ๐Ÿ“ƒ
https://www.assemblyai.com/playground/transcript/64kyzev80o-6ed4-4902-a066-7df25c363193

Andre Karpathi is a founding member of OpenAI. He will talk about how we train GPT assistants. In the second part he will take a look at how we can use these assistants effectively for your applications.

TRAINING NEURAL NETWORKS ON THE INTERNET

We have four major stages pretraining supervised fine tuning, reward modeling, reinforcement learning. In each stage we have a data set that powers that stage. And then we have an algorithm that for our purposes will be an objective for training a neural network.

GPT 3.1: BASE MODELS AND AGENTS

The GPT four model that you might be interacting with over API is not a base model, it's an assistant model. You can even trick base models into being assistants. Instead we have a different path to make actual GPT assistance, not just base model document completers.

NEUROANATOMY 2.8

In the reward modeling step, what we're going to do is we're now going to shift our data collection to be of the form of comparisons. Now, because we have a reward model, we can score the quality of any arbitrary completion for any given prompt. And then at the end, you could deploy a Rlhf model.

COGNITIVE PROCESSES AND GPT

How do we best apply a GPT assistant model to your problems? Think about the rich internal monologue and tool use and how much work actually goes computationally in your brain to generate this one final sentence. From GPT's perspective, this is just a sequence of tokens.

TREE OF THOUGHT AND PROMPT ENGINEERING

A lot of people are really playing around with kind of prompt engineering to bring back some of these abilities that we sort of have in our brain for LLMs. I think this is kind of an equivalent of AlphaGo but for text. I would not advise people to use it in practical applications.

WHAT ARE THE QUIRKS OF LLMS?

The next thing that I find kind of interesting is that LLMs don't want to succeed, they want to imitate. And so at test time, you actually have to ask for a good performance. Next up, I think a lot of people are really interested in basically retrieval augmented generation.

CONSTRAINT PROMPTING IN LLMS

Next, I wanted to briefly talk about constraint prompting. This is basically techniques for forcing a certain template in the outputs of LLMs. And I think this kind of constraint sampling is also extremely interesting.

FINE-TUNING A LANGUAGE MODEL

You can get really far with prompt engineering, but it's also possible to think about fine tuning your models. Fine tuning is a lot more technically involved. It requires human data contractors for data sets and or synthetic data pipelines. Break up your task into two major parts.

LIMITS TO FULLY AUTONOMOUS LLMS

There's a large number of limitations to LLMs today, so I would keep that definitely in mind for all your applications models. My recommendation right now is use LLMs in low stakes applications, combine them with always with human oversight. Think copilots instead of completely autonomous agents.
๐Ÿง‘๐Ÿผโ€โœˆ๏ธ ๐Ÿšง๐Ÿ’ป
In this post, I try to answer specific questions about the internals of Copilot, while also describing some interesting observations I made as I combed through the code. I will provide pointers to the relevant code for almost everything I talk about, so that interested folks can take a look at the code themselves.

https://thakkarparth007.github.io/copilot-explorer/posts/copilot-internals
<์ž๊ทน์„ ์ค„์ด๊ณ  ์ƒ๊ฐ์„ ๋Š˜๋ฆฌ๊ธฐ>

์š”์ฆ˜ ํ˜„๋Œ€์ธ๋“ค์€ ๊ฑฐ์˜ ADHD ์ƒํƒœ๋กœ ์ผ์„ ํ•œ๋‹ค๊ณ  ์ƒ๊ฐ์ด ๋“œ๋Š” ๋ฉด์ด ์žˆ๋‹ค. ์ง€์†์ ์œผ๋กœ ๋†’์€ ๊ฐ•๋„์˜ ์ž๊ทน์— ์ž์‹ ์„ ๋…ธ์ถœ์‹œํ‚ค๊ธฐ ์‰ฝ๊ธฐ ๋•Œ๋ฌธ์ด๋‹ค. ์ด๋Ÿฐ ํ™˜๊ฒฝ์†์—์„œ ๋ญ ํ•˜๋‚˜์— ์ฐจ๋ถ„ํ•˜๊ฒŒ ์ง‘์ค‘ํ•˜๊ณ  ๊นŠ์ด ์žˆ๋Š” ์‚ฌ๊ณ ๋ฅผ ํ•˜๊ธฐ๊ฐ€ ํž˜๋“ค๋‹ค.

๋‘ ๊ฐ€์ง€ ์‚ฌ๋ก€๋ฅผ ๋จผ์ € ์†Œ๊ฐœํ•˜๊ฒ ๋‹ค.

์‚ฌ๋ก€ 1)

๋‚ด๊ฐ€ ์•„๋Š” K๋ชจ์”จ๋Š” ๋Œ€๊ธฐ์—… ์ง์›์ด์—ˆ๋Š”๋ฐ, ํ•˜๋ฃจ์— ์ „์‚ฌ์—์„œ ๋“ค์–ด์˜ค๋Š” ์—…๋ฌด ์š”์ฒญ๋งŒ ์ˆ˜๋ฐฑ๊ฑด์ด๋ผ๊ณ  ํ–ˆ๋‹ค. ๊ทธ๋ž˜์„œ ๋‚ ๋งˆ๋‹ค ๋ฐค 11์‹œ์— ํ‡ด๊ทผ์„ ํ•˜๊ณ  ์žˆ์—ˆ๋‹ค.

๊ทธ๋Ÿฌ๋‹ค๊ฐ€ ๋‚˜์—๊ฒŒ์„œ ์• ์ž์ผ ์ด์•ผ๊ธฐ๋ฅผ ๋“ฃ๊ณ  ์‹คํ—˜์„ ํ•ด๋ณด๊ธฐ๋กœ ๊ฒฐ์‹ฌํ–ˆ๋‹ค. ์ •์‹œ ํ‡ด๊ทผ. ๊ทธ๋ž˜์„œ ํŒ€์žฅ์—๊ฒŒ ์ œ์•ˆ์„ ํ–ˆ๋‹ค. ์˜ค๋Š˜๋ถ€ํ„ฐ 18์‹œ ์ •์‹œ ํ‡ด๊ทผ์„ ํ•˜๊ฒ ๋‹ค. ํ˜น์—ฌ ์ผ ์ฒ˜๋ฆฌ๊ฐ€ ์กฐ๊ธˆ์ด๋ผ๋„ ๋–จ์–ด์ง„๋‹ค๋Š” ๋А๋‚Œ์ด ๋“ค๋ฉด ์–˜๊ธฐํ•ด๋ผ. ๋ฐ”๋กœ ์›๋ณตํ•˜๊ฒ ๋‹ค. ๊ทธ๋Ÿฌ๊ณ  ๊ทธ๋‚ ๋ถ€ํ„ฐ 18์‹œ ํ‡ด๊ทผ์„ ํ–ˆ๋‹ค. ์ง‘์— ์˜ค๋ฉด ์ €๋… 7์‹œ๋ถ€ํ„ฐ 9์‹œ๊นŒ์ง€ ๋‘ ์‹œ๊ฐ„์”ฉ 6์‚ด ์•„์ด๋ž‘ ๋†€์•„์คฌ๋‹ค๊ณ  ํ•œ๋‹ค. ๊ทธ์ „๊นŒ์ง€ ์•„์ด์—๊ฒŒ ์•„๋น ๋Š” ์—†๋Š” ์กด์žฌ์˜€๋‹ค. ์ฃผ์ค‘์—๋Š” ๋ฐค 11์‹œ์— ์˜ค๊ณ , ์•„์นจ์—๋Š” ์ž๊ธฐ๋ณด๋‹ค ๋จผ์ € ๋‚˜๊ฐ€๊ณ  ์ฃผ๋ง์—๋Š” ๊ณ„์† ์“ฐ๋Ÿฌ์ ธ ์žˆ์—ˆ์œผ๋‹ˆ. ๊ทธ๋Ÿฐ ์•„์ด์—๊ฒŒ "์•„๋น "๊ฐ€ ์ƒ๊ธด ๊ฑฐ๋‹ค.

๊ทผ๋ฐ ๋ฌธ์ œ๊ฐ€ ํ•˜๋‚˜ ์žˆ์—ˆ๋‹ค. ์ •์‹œ ํ‡ด๊ทผ์„ ํ–ˆ์œผ๋‹ˆ ๋‹ค ์ฒ˜๋ฆฌ ๋ชปํ•œ ์ผ๋“ค์ด ๋ฌธ์ œ. ๊ทธ๋Ÿฐ๋ฐ ๋ณด์•ˆ๋ฌธ์ œ ๋•Œ๋ฌธ์— ์ง‘์—์„œ ํšŒ์‚ฌ ์ปดํ“จํ„ฐ๋‚˜ ์ž๋ฃŒ์— ์ ‘๊ทผํ•  ์ˆ˜๊ฐ€ ์—†์—ˆ๋‹ค. ๊ทธ๋ž˜์„œ ๊ทธ๊ฐ€ ๋Œ€์•ˆ์œผ๋กœ ํ–ˆ๋˜ ๊ฑฐ๋Š” ๋ฐค 11์‹œ๋ถ€ํ„ฐ 1์‹œ๊นŒ์ง€ ๋‘ ์‹œ๊ฐ„ ๋™์•ˆ ์ž๊ธฐ ์ฑ…์ƒ์— ์ด๋ฉด์ง€ ํŽผ์น˜๊ณ  ์•‰์•„์„œ ์˜ค๋Š˜ ํ–ˆ๋˜ ์ผ๋“ค, ๋‚ด์ผ ํ•  ์ผ๋“ค์„ ์–ด๋–ป๊ฒŒ ํ•ด์•ผ ๋” ํ˜„๋ช…ํ•˜๊ฒŒ ์ฒ˜๋ฆฌํ•  ๊ฑด๊ฐ€ ์ „๋žต์„ ์งœ๋Š” ๊ฑฐ์˜€๋‹ค. ๊ทธ๊ฑธ ๋‚ ๋งˆ๋‹ค ํ–ˆ๋‹ค.

๊ทธ๋Ÿฌ๊ณ  ๋‹ค์Œ๋‚  ์ถœ๊ทผ์„ ํ•˜๋‹ˆ ์—…๋ฌด ์š”์ฒญ ์ค‘์˜ 50% ์ด์ƒ์€ ์ž๋™์œผ๋กœ ํ•ด๊ฒฐ๋œ ๊ฒฝ์šฐ๊ฐ€ ๋งŽ์•˜๊ณ (์š”์ฒญํ•œ ๋ถ€์„œ์—์„œ ๋‹ต๋‹ตํ•˜๋‹ˆ ์ž์ฒด์ ์œผ๋กœ ํ•ด๊ฒฐ), ๋‚จ์€ 50%๋Š” ์ง€๋‚œ ๋ฐค์— ๊ณ ๋ฏผํ•œ ๊ฒฐ๊ณผ ๋” ํ˜„๋ช…ํ•œ ๋ฐฉ๋ฒ•์œผ๋กœ ์ฒ˜๋ฆฌ๋ฅผ ํ•ด์„œ ๊ธˆ๋ฐฉ ๋๋‚ผ ์ˆ˜ ์žˆ์—ˆ๋‹ค.

๋ฌผ๋ก  ๊ฒฐ๊ณผ์ ์œผ๋กœ ๋ฐค 11์‹œ ํ‡ด๊ทผํ•  ๋•Œ๋ณด๋‹ค ์ˆ˜๋ฉด์‹œ๊ฐ„์ด ์ค„์—ˆ๋‹ค๊ณ  ํ•œ๋‹ค. ์˜ˆ์ „์—๋Š” ์ง‘์— ๋“ค์–ด์˜ค๋ฉด ๋ฐ”๋กœ ์“ฐ๋Ÿฌ์ ธ์„œ ์žค์œผ๋‹ˆ๊นŒ. ํ•˜์ง€๋งŒ ๋ชธ์ด ๋А๋ผ๋Š” ์—๋„ˆ์ง€๋Š” ํ›จ์”ฌ ์ข‹์•„์กŒ๋‹ค๊ณ  ํ•œ๋‹ค.

์‚ฌ๋ก€ 2)

์˜ˆ์ „์— ๊ตฐ๋Œ€์‹œ์ ˆ ์ž๋Œ€ ๋ฐฐ์น˜๋ฅผ ๋ฐ›๊ณ  ํ•ด๋‹น ๋ถ€๋Œ€์— ๊ฐ”๊ณ  ์‚ฌ์ˆ˜๋ฅผ ๋ฐฐ๋‹น ๋ฐ›์•˜๋‹ค. ๊ทผ๋ฐ ๊ทธ ์‚ฌ์ˆ˜ ์–ผ๊ตด์„ ๋ณด๊ธฐ๊ฐ€ ํž˜๋“ ๊ฑฐ๋‹ค. ๋ฉฐ์น  ์ง€๋‚˜ ์•Œ๊ฒŒ ๋๋Š”๋ฐ ๊ทธ ์‚ฌ์ˆ˜ ์ „์—ญ์ผ์ด 1์ฃผ์ผ ๋’ค๋ž€๋‹ค. ๋‚ด ์‚ฌ์ˆ˜์˜ ๋ณด์ง์€ ๋Œ€๋Œ€ ์ •๋น„๊ณผ ์„œ๋ฌด๋ณ‘. ์›Œ๋‚™ ํ•˜๋Š” ์ผ์ด ๋งŽ๊ณ  ๋ณต์žกํ•ด์„œ ํ†ต์ƒ 1๋…„ ์ •๋„๋Š” ์ธ์ˆ˜์ธ๊ณ„๋ฅผ ๋ฐ›์•„์•ผ ์ œ๋Œ€๋กœ ์ผ์„ ํ•˜๊ฒŒ ๋œ๋‹ค๊ณ  ํ•œ๋‹ค. ๊ทผ๋ฐ ์ด ์‚ฌ๋žŒ์€ 1์ฃผ์ผ ๋’ค์— ์ „์—ญํ•˜๊ณ , ์ด 1์ฃผ์ผ๋„ ์–ผ๋ ๋šฑ๋•… ์ง€๋‚˜๊ฐ€๊ณ  ์žˆ์—ˆ๋‹ค. ๊ฐ€๋” ์ •๋น„๊ณผ์— ๋‚ด๋ ค์™€์„œ๋Š” ๊ถ๊ธˆํ•œ ๊ฑฐ ๋ฌผ์–ด๋ดํ•˜๊ณ  ๋ˆ„์›Œ์žˆ๊ฑฐ๋‚˜ ํ•˜๋Š” ์ •๋„. ์ •๋ง ๋ฌธ์ œ๋Š” ์ด ์‚ฌ๋žŒ์˜ ๋ณด์ง์„ ์ •ํ™•ํ•˜๊ฒŒ ํŒŒ์•…ํ•˜๋Š” ์‚ฌ๋žŒ์ด ๊ฐ„๋ถ€๋‚˜ ๋ณ‘ ์ค‘์— ์•„๋ฌด๋„ ์—†๋‹ค๋Š” ๊ฑฐ.

๊ฒฐ๊ตญ ๋‚˜๋Š” ๊ฑฐ์˜ ์•„๋ฌด๊ฒƒ๋„ ๋ฐฐ์šฐ์ง€๋„ ๋ชปํ•œ ์ฑ„๋กœ ์‚ฌ์ˆ˜๊ฐ€ ์ „์—ญ์„ ํ–ˆ๊ณ , ์—…๋ฌด ๋งค๋‰ด์–ผ๋„ ํ•˜๋‚˜ ์—†์—ˆ๋‹ค. ์ฐธ๊ณ ํ•  ์ž๋ฃŒ๊ฐ€ ์ „ํ˜€ ์—†๋Š” ์ƒํ™ฉ.

๊ณ ๋ฏผํ•˜๋‹ค๊ฐ€ ๊ฒฐ๊ตญ ํ•˜๊ฒŒ ๋œ ์„ ํƒ์€ ์›๋ฆฌ์™€ ์›์น™์œผ๋กœ ์ƒ๊ฐํ•ด์„œ ํ–‰๋™ํ•˜์ž๋Š” ๊ฑฐ์˜€๋‹ค. ์–ด๋–ค ๋ฌธ์ œ ์ƒํ™ฉ์ด ๋ฐœ์ƒํ•˜๋ฉด ๋‚ด๊ฐ€ ์ƒ๊ฐํ•˜๋Š” ๊ธฐ๋ณธ์ ์ธ ์›๋ฆฌ์— ๋”ฐ๋ผ(์˜ˆ์ปจ๋Œ€ ์–ด๋–ป๊ฒŒ ํ•˜๋Š” ๊ฒƒ์ด ์œก๊ตฐ์—๊ฒŒ ์ด๋“์ด ๋˜๋Š” ํ–‰๋™์ธ๊ฐ€ ๊ฐ™์€) ๋…ผ๋ฆฌ์ ์œผ๋กœ ๋ง์ด ๋˜๋Š” ํ–‰๋™์„ ์ƒ๊ฐํ•ด์„œ ํ–ˆ๋‹ค. ๋‚ด๊ฐ€ ๋ชจ๋“  ๊ทœ์น™๊ณผ ๋ฒ•์„ ์„ค๊ณ„ํ•˜๋ฉด์„œ ํ–ˆ๋‹ค๊ณ  ํ• ๊นŒ. ์ด๋Ÿฌ๋‹ˆ๊นŒ ๊ฑฐ์น  ๊ฒƒ์ด ์—†์—ˆ๋‹ค. ๋ญ๋“ ์ง€ ๊นŠ๊ฒŒ ์ƒ๊ฐํ•ด์„œ ๊ทธ๋Œ€๋กœ ํ•˜๋ฉด ๋‹ค ํ’€๋ฆฌ๋”๋ผ๋Š”.

๊ทผ๋ฐ ์˜์™ธ๋กœ ์ด ๋ฐฉ๋ฒ•์ด ์ž˜ ํ†ตํ–ˆ๋‹ค. ๊ทธ๋ž˜์„œ ๊ฒฐ๊ตญ ๋‚ด๊ฐ€ ๋ชจ๋“  ์ฒด๊ณ„๋ฅผ ๋งŒ๋“ค์—ˆ๊ณ  ์ด๊ฑธ๋กœ ์ƒ๋„ ๋ช‡๋ฒˆ ๋ฐ›์•˜๋‹ค. ๊ตฐ๋‹จ์—์„œ ๊ฐ์‚ฌ ๋‚ด๋ ค์™”์„ ๋•Œ์—๋Š” ๋‚ด๊ฐ€ ๊ตฐ๋ฌด์›์ด๋ž‘ ์žฅ๊ต๋“ค ๋ชจ์•„๋†“๊ณ  ๋น„๊ณต์‹ ๊ฐ•์—ฐ๋„ ํ–ˆ๋‹ค.

----
๋•Œ๋กœ๋Š” ์™ธ๋ถ€ ์ž๊ทน/์ •๋ณด๋ฅผ ์ œํ•œํ•˜๊ณ  ์ƒ๊ฐ์— ์ง‘์ค‘ํ•˜๋Š” ๊ฒƒ์ด ๋„์›€์ด ๋˜๋Š” ๊ฒฝ์šฐ๊ฐ€ ์žˆ๋‹ค. ๋ค์œผ๋กœ ์ƒ๊ฐํ•˜๋Š” ๊ทผ์œก๊ณผ ๊ธฐ์ˆ ๋„ ๋Š˜๊ฒŒ ๋œ๋‹ค.

๊ทธ๋ž˜์„œ ๋‚˜๋Š” ์˜ˆ์ปจ๋Œ€ ๋‹ค์Œ๊ณผ ๊ฐ™์€ ๊ฒƒ๋“ค์„ ์ถ”์ฒœํ•œ๋‹ค:
* ๋ฒ„๊ทธ๊ฐ€ ๋‚˜์˜ค๋ฉด ๋ฐ”๋กœ ๊ฒ€์ƒ‰์ฐฝ์— ๋•Œ๋ ค๋„ฃ์ง€ ๋ง๊ณ  ์ ์–ด๋„ 5๋ถ„, 10๋ถ„๊ฐ„์€ ๋ฐฑ์ง€์—๋‹ค๊ฐ€ ๋ฌธ์ œ์ƒํ™ฉ์„ ๊ทธ๋ ค๋ณด๊ณ  ์›์ธ ์œ ์ถ”ํ•ด๋ณด๊ธฐ
* ์ „ํ˜€ ๋ชจ๋ฅด๋Š” ๋ถ„์•ผ์— ์ž…๋ฌธํ•˜๊ณ  ์‹ถ์„ ๋•Œ ์ธํ„ฐ๋„ท ๊ฒ€์ƒ‰๋ณด๋‹ค๋Š” ์„œ์ ์—์„œ ์ž˜๋‚˜๊ฐ€๋Š” ์ฑ… ์ค‘์— ์Šคํƒ€์ผ์ด ๋‹ค๋ฅธ ์ฑ… 3๊ถŒ์„ ๊ตฌ์ž…ํ•ด์„œ ์–˜๋ฅผ ๋น„๊ตํ•ด๋ณด๋ฉด์„œ ๋ณด๊ธฐ (๋‚˜๋Š” ์ด๊ฑธ bounded exploration์ด๋ผ๊ณ  ๋ถ€๋ฅธ๋‹ค -- ์ด๊ฑธ ์•ˆํ•˜๋ฉด ์–ด๋А ํ•˜๋‚˜ ์ œ๋Œ€๋กœ ๋ณด์ง€ ์•Š๊ณ  ๊ณ„์† ๊น”์ง๊น”์ง ๋Œ€๋ฉด์„œ ์‹œ๊ฐ„์„ ๋‚ญ๋น„ํ•˜๊ธฐ ์‰ฝ๋‹ค)
* ํ•ด๊ฒฐํ•ด์•ผํ•  ๋ณต์žกํ•œ ๋ฌธ์ œ๊ฐ€ ์žˆ์„ ๊ฒฝ์šฐ ์ถ”๊ฐ€ ์ •๋ณด๋ฅผ ์ „ํ˜€ ์ฐพ์ง€ ์•Š๊ณ  ๋ฐฑ์ง€๋ฅผ ํŽผ์ณ๋†“๊ณ  30๋ถ„ ๋™์•ˆ ๋…ผ๋ฆฌ์™€ ๋‚ด ์ƒ๊ฐ, ๋‚ด ๊ณผ๊ฑฐ๊ฒฝํ—˜์œผ๋กœ๋งŒ ํ•ด๊ฒฐ์ฑ…์„ ์„ค๊ณ„ํ•ด ๋ณด๊ธฐ

https://www.facebook.com/100000557305988/posts/pfbid02joCFDgeyR58vuv2MyZqQWJ1cf7FwrYZHS6FLq9ox8Bqu2RE9cV3HdgzWdHJvopjkl/?mibextid=jf9HGS
๐Ÿ‘5
Continuous Learning_Startup & Investment
Could one Language Learning Model handle all programming languages? Or should we tailor a model for each? What's your take? #LLM #ProgrammingLanguages https://www.linkedin.com/posts/mateizaharia_introducing-english-as-the-new-programming-activity-7080242815120637952โ€ฆ
๋„ˆ๋ฌด๋‚˜ ์‰ฌ์›Œ์ง€๋Š” ๋ฐ์ดํ„ฐ ์‚ฌ์ด์–ธ์Šค ๐Ÿš€

ChatGPT ๋•๋ถ„์— ๋ฐ์ดํ„ฐ ์‚ฌ์ด์–ธ์Šค๊ฐ€ ๋†€๋ž๋„๋ก ์‰ฌ์›Œ์ง€๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ๐Ÿค– ์ด์ „์—๋Š” ํด๋Ÿฌ์Šคํ„ฐ๋ง์„ ์ด์šฉํ•œ ์•„๋ž˜ ์ฐจํŠธ๋ฅผ ๋งŒ๋“ค๊ธฐ ์œ„ํ•ด ํ•„์š”ํ–ˆ๋˜ ์ง€์‹๋“ค์€ ๋‹ค์Œ๊ณผ ๊ฐ™์•˜์Šต๋‹ˆ๋‹ค.

## Google Colab ํ•™์Šต ์‹œ๊ฐ„ ๐Ÿ“š:

1. ๊ธฐ๋ณธ์ ์ธ ์‚ฌ์šฉ๋ฒ•์„ ์ตํžˆ๋Š”๋ฐ ์•ฝ 1์ฃผ ์ •๋„์˜ ํ•™์Šต ์‹œ๊ฐ„์ด ํ•„์š”ํ–ˆ์Šต๋‹ˆ๋‹ค.

2. ๋” ๋ณต์žกํ•œ ์ž‘์—…, ์˜ˆ๋ฅผ ๋“ค์–ด ์™ธ๋ถ€ ๋ฐ์ดํ„ฐ๋ฅผ ๋ถˆ๋Ÿฌ์˜ค๊ฑฐ๋‚˜, ํฐ ๊ทœ๋ชจ์˜ ๋ฐ์ดํ„ฐ๋ฅผ ์ฒ˜๋ฆฌํ•˜๋Š” ๋ฐฉ๋ฒ• ๋“ฑ์„ ํ•™์Šตํ•˜๋Š”๋ฐ ์ถ”๊ฐ€์ ์ธ 1~2์ฃผ์˜ ์‹œ๊ฐ„์ด ํ•„์š”ํ–ˆ์Šต๋‹ˆ๋‹ค.

## ๋ฐ์ดํ„ฐ ๊ณผํ•™ ๋ฐฐ๊ฒฝ ์ง€์‹ ๐ŸŽ“:

1. ํด๋Ÿฌ์Šคํ„ฐ๋ง: ๊ธฐ๋ณธ์ ์ธ ์ดํ•ด๋ฅผ ์œ„ํ•ด 1~2์ฃผ์˜ ํ•™์Šต ์‹œ๊ฐ„์ด ํ•„์š”ํ–ˆ์Šต๋‹ˆ๋‹ค.

2. ํด๋Ÿฌ์Šคํ„ฐ๋ง ํ‰๊ฐ€ ์ง€ํ‘œ: ๊ฐ ์ง€ํ‘œ์— ๋Œ€ํ•œ ๊ธฐ๋ณธ์ ์ธ ์ดํ•ด๋ฅผ ์œ„ํ•ด 1์ฃผ ์ •๋„์˜ ํ•™์Šต ์‹œ๊ฐ„์ด ํ•„์š”ํ–ˆ์Šต๋‹ˆ๋‹ค.

3. ๋ฐ์ดํ„ฐ ๋ถ„์„ ๋ฐ ์ฒ˜๋ฆฌ: ์ด ์ฃผ์ œ๋Š” ๊ด‘๋ฒ”์œ„ํ•˜๋ฏ€๋กœ, ๊ธฐ๋ณธ์ ์ธ ๋ฐ์ดํ„ฐ ์ „์ฒ˜๋ฆฌ ๋ฐ ๋ถ„์„ ๊ธฐ๋ฒ•์„ ์Šต๋“ํ•˜๋Š” ๋ฐ๋Š” ์ตœ์†Œํ•œ 1~2๊ฐœ์›”์˜ ํ•™์Šต ์‹œ๊ฐ„์ด ํ•„์š”ํ–ˆ์Šต๋‹ˆ๋‹ค.

## API ์ง€์‹ ๐Ÿ’ป:

1. Firebase Firestore: Firestore์˜ ๊ธฐ๋ณธ์ ์ธ ์‚ฌ์šฉ๋ฒ•์„ ๋ฐฐ์šฐ๋Š” ๋ฐ๋Š” 1~2์ฃผ์˜ ์‹œ๊ฐ„์ด ์†Œ์š”๋์Šต๋‹ˆ๋‹ค.

## ์ฝ”๋”ฉ ์Šคํ‚ฌ ๐Ÿ–ฅ๏ธ:

1. ํŒŒ์ด์ฌ: ํŒŒ์ด์ฌ์˜ ๊ธฐ๋ณธ ๋ฌธ๋ฒ•์„ ์ตํžˆ๋Š” ๋ฐ๋Š” ์•ฝ 1~2๊ฐœ์›”์˜ ํ•™์Šต ์‹œ๊ฐ„์ด ํ•„์š”ํ–ˆ์Šต๋‹ˆ๋‹ค.

2. NumPy: ๊ธฐ๋ณธ์ ์ธ NumPy ๊ธฐ๋Šฅ์„ ์ตํžˆ๋Š” ๋ฐ๋Š” ์•ฝ 1~2์ฃผ์˜ ํ•™์Šต ์‹œ๊ฐ„์ด ํ•„์š”ํ–ˆ์Šต๋‹ˆ๋‹ค.

3. Matplotlib: ๊ธฐ๋ณธ์ ์ธ ๊ทธ๋ž˜ํ”„๋ฅผ ๊ทธ๋ฆฌ๋Š” ๋ฐฉ๋ฒ•์„ ๋ฐฐ์šฐ๋Š” ๋ฐ๋Š” ์•ฝ 1์ฃผ์˜ ํ•™์Šต ์‹œ๊ฐ„์ด ํ•„์š”ํ–ˆ์Šต๋‹ˆ๋‹ค.

์œ„์—์„œ ์ œ์‹œํ•œ ๊ฐ ํ•ญ๋ชฉ์˜ ํ•™์Šต ์‹œ๊ฐ„์„ ํ•ฉ์‚ฐํ•˜๋ฉด ๋Œ€๋žต์ ์œผ๋กœ ๋‹ค์Œ๊ณผ ๊ฐ™์Šต๋‹ˆ๋‹ค: ๋ฐ์ดํ„ฐ ๊ณผํ•™ ๊ธฐ์ดˆ: ์•ฝ 2~4๊ฐœ์›”, API ์ง€์‹ (Firebase Firestore): ์•ฝ 1~2์ฃผ, ์ฝ”๋”ฉ ์Šคํ‚ฌ (ํŒŒ์ด์ฌ, NumPy, Matplotlib): ์•ฝ 2~3๊ฐœ์›”. ๋”ฐ๋ผ์„œ ์ด ํ•™์Šต ์‹œ๊ฐ„์€ ์•ฝ 4~7๊ฐœ์›” ์ •๋„๋กœ ์˜ˆ์ƒํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๐Ÿ“ˆ

----

# ChatGPT๋ฅผ ์ด์šฉํ•˜๋‹ˆ ๋‹ค์Œ๊ณผ ๊ฐ™์ด ๋˜์–ด๋ฒ„๋ ธ์Šต๋‹ˆ๋‹ค. ๐Ÿ”„

AI๊ฐ€ ์ฝ”๋”ฉ๊ณผ ์‹คํ—˜ ์„ค๊ณ„๋ฅผ ๋‹ด๋‹นํ•˜๋ฏ€๋กœ ๊ทธ ๋ถ€๋ถ„์˜ ํ•™์Šต ์‹œ๊ฐ„์€ ์ œ์™ธํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋ฏ€๋กœ, ๋‚จ์€ ๋ถ€๋ถ„์€ ๋ฐ์ดํ„ฐ ๊ณผํ•™์— ๋Œ€ํ•œ ๊ฐ€๋ฒผ์šด ๋ฐฐ๊ฒฝ ์ง€์‹๊ณผ Google Colab์— ๋Œ€ํ•œ ์ดํ•ด์ž…๋‹ˆ๋‹ค. ๐Ÿค”

1. ๋ฐ์ดํ„ฐ ๊ณผํ•™ ๋ฐฐ๊ฒฝ ์ง€์‹: AI ๋น„์„œ์˜ ์„ค๋ช…๊ณผ ๊ฐ€์ด๋“œ๋กœ, ์•ฝ 1๊ฐœ์›”๋กœ ๋‹จ์ถ•๋  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๊ฒฝ์šฐ์— ๋”ฐ๋ผ์„œ๋Š” 2์ฃผ์—๋„ ๊ธฐ๋ณธ ๊ฐœ๋…์„ ํ›‘์„ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.

2. Google Colab: AI ๋น„์„œ์˜ ๋„์›€์œผ๋กœ, ํ•™์Šต ์‹œ๊ฐ„์„ ์•ฝ 1์ฃผ๋กœ ์ค„์ผ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. - ์‚ฌ์‹ค 1์‹œ๊ฐ„๋งŒ ํ•ด๋„ ๋  ๊ฒƒ ๊ฐ™๊ธด ํ•ด์š”๏ฟผ

์ด ๊ฒฝ์šฐ, ์ด ํ•™์Šต ์‹œ๊ฐ„์€ ์•ฝ 1~2๊ฐœ์›” ์ •๋„๋กœ ์ถ”์ •๋ฉ๋‹ˆ๋‹ค. ์ด๋ฏธ ์ฝ”๋”ฉ ์Šคํ‚ฌ๊ณผ API ์‚ฌ์šฉ์— ๋Œ€ํ•œ ์ง€์‹์ด ์žˆ๋‹ค๋ฉด, ์ด ์‹œ๊ฐ„์€ ๋”์šฑ ๋‹จ์ถ•๋  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. โŒ›

----

๊ฒฐ๊ตญ ์ดˆ๋ณด์ž์˜ ๊ฒฝ์šฐ 6๊ฐœ์›” ์ฝ”์Šค -> 1๊ฐœ์›” ์ฝ”์Šค๊ฐ€ ๋ฉ๋‹ˆ๋‹ค. ๐ŸŽ‰ ๋ฐ์ดํ„ฐ ์‚ฌ์ด์–ธ์Šค ๋ฐฐ๊ฒฝ ์ง€์‹์„ ์•Œ๊ณ  ์žˆ๊ณ  ํŒŒ์ด์ฌ ๋ผ์ด๋ธŒ๋Ÿฌ๋ฆฌ ์‚ฌ์šฉ ๋ฐฉ๋ฒ•์„ ๋ชฐ๋ž๋˜ ์ œ ์ž…์žฅ์—์„œ๋Š” 3์ฃผ ์ •๋„์—์„œ ๋‘์‹œ๊ฐ„์œผ๋กœ ๋‹จ์ถ• ๋˜์—ˆ์Šต๋‹ˆ๋‹ค. ๐Ÿ˜ฒ ์ด ์™ธ์—๋„ ๋ฐ์ดํ„ฐ ๊ณผํ•™ ์ „๋ฐ˜์„ ๋ฐฐ์šฐ๋ ค๋ฉด 4๋…„๋„ ๋ชจ์ž๋ž๋‹ˆ๋‹ค.

๊ฒฐ๊ตญ ์‹œ๋‹ˆ์–ด ๋ฐ์ดํ„ฐ ์‚ฌ์ด์–ธํ‹ฐ์Šค ํ•œ๋ช…์ด ํ•  ์ˆ˜ ์žˆ๋Š” ์ผ์ด ์ฅฌ๋‹ˆ์–ด ์‚ฌ์ด์–ธํ‹ฐ์ŠคํŠธ์™€ ์ฅฌ๋‹ˆ์–ด ๋ฐ์ดํ„ฐ ์—”์ง€๋‹ˆ์–ด 10๋ช… ์ด์ƒ์— ํ•ด๋‹นํ•˜๋Š” ์ผ์ด ๋˜์–ด๋ฒ„๋ฆฝ๋‹ˆ๋‹ค.

์‹ค๋ฆฌ์ฝ˜๋ฐธ๋ฆฌ์—์„œ๋Š” ์ด๋ฏธ ์ฅฌ๋‹ˆ์–ด ๋ฐ์ดํ„ฐ ๊ณผํ•™์ž๋“ค์ด ๋น ๋ฅธ ์†๋„๋กœ ์ง์—…์„ ์žƒ๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ๐Ÿ˜ฑ

ํ•™๊ต์—์„œ์˜ ๊ณผ์ •๋„ ๋ฐ”๋€Œ์–ด์•ผ ํ•  ๊ฒƒ ๊ฐ™์Šต๋‹ˆ๋‹ค. ์˜คํžˆ๋ ค ๊ฐ™์€ ์‹œ๊ฐ„ ๋‚ด์— ๋” ๊นŠ์ด ์žˆ๋Š” ์ด๋ก ์„ ๋ฐฐ์šธ ์ˆ˜ ์žˆ์„ ๊ฒƒ ๊ฐ™์Šต๋‹ˆ๋‹ค. ๋˜ํ•œ ์‹ค์ œ ์ฝ”๋”ฉ๋ณด๋‹ค๋Š” ์—ฐ๊ตฌ ๋ฐฉ๋ฒ•๋ก ์— ์ค‘์ ์„ ๋‘๊ณ  ๊ต์œก ์„ค๊ณ„๋ฅผ ํ•ด์•ผ ํ•  ๊ฒƒ ๊ฐ™์Šต๋‹ˆ๋‹ค. ๋ฐ์ดํ„ฐ ์‚ฌ์ด์–ธํ‹ฐ์ŠคํŠธ๋“ค์ด ์‹ค๋ฌด ๊ธฐ์ˆ ๋ณด๋‹ค ์ง€์‹์ ์œผ๋กœ ์ƒํ–ฅ ํ‰์ค€ํ™” ๋˜๋Š” ์ƒํ™ฉ์ด ์˜ฌ ๊ฒƒ ๊ฐ™์Šต๋‹ˆ๋‹ค.

---

์•„๋ž˜ scatter plot์„ ์œ„ํ•ด ์‚ฌ์šฉํ•œ prompt:

1. Get the latest 1000 samples from user_tribes collection
2, tribeId is cluster id, x and y are the coordinates.
3. Measure the homogeneity and completeness using colab.
4. Visualize the results.

Kmeans๋ผ๊ณ  ๋ง๋„ ์•ˆ ํ–ˆ๋Š”๋ฐ ์•Œ์•„์„œ ๊ฐ–๋‹ค ์“ฐ๋„ค์š”.

https://www.facebook.com/634740022/posts/pfbid0cuABUXxgECdMwZfQaZ9u88HqXaLoLKzdJxBGLSsfHMfUovKRdQnuybjUYc9sJycsl/?mibextid=jf9HGS