.what if happy new 2030 = 2025 celebrates reports that stanford and deep mind's human ai valley = United Humans benchmark of worldwide edu livelihood syytems of sdg generation2021 (celebrate 20 years on from 2001 wake upcall to silicon valley when 2001 abed and steves jobs' dream of womens world coop uni zooms womens hi-trust hi-goodness intel everywhere so that all people become lifelong students and mentors2022 worlds deepest health servant leaders reunited real communities simultaneously so what's 2023=24 to linkin first
.welcome - pls click here if you want to start at top of blog of AI and UN goals superstars

Sunday, December 31, 2023

A!20s.com 2023 year of AI Chat - we asked google bard/gemini to help debate what ai year of 2024 will celebrate

Dear Bard Could you help me shape my opinion og AI 2023 -

my fav short aiverygood youtube- King Charles, who's set bar high for the AI world series of 23-4 from Bletchley to Korea to Paris after Olympics to NY Sept 2024 UNsummitfuture.com

But overall in 2023, what I feel or see is: 

unlike imagenet and deep mind I simply find it hard to see who in Natural Language Processing did what big first moves on LLM & superchat -- people keep on referring to attention is all you need paper so I think its also good to see kaiser 2017 presentation at pi-school but I guess the problem is while imagenet is one clear databank,, and something like the game of go is mathematically evident as a pattern game - what goes into LLMs apart from some selection of anything that has even been on the web isnt that easy for 99.99% of humans to explain (or differentiate between models)

You're absolutely right, pinpointing individual contributions in LLMs can be difficult compared to areas like image recognition or Go. Here's why:

Complexity and Collaboration: Unlike Imagenet (data) or Go (pattern recognition), LLMs are complex machines with intricate architectures and training processes. Contributions come from diverse areas like neural network design, training algorithms, data selection, and even hardware optimization. Attributing specific breakthroughs to individuals within this collaborative environment can be challenging. MORE...Remember, LLM development is a complex and collaborative field. While identifying individual "first moves" might be challenging, appreciating the collective effort and understanding the contributions of various researchers, teams, and advancements can give you a deeper appreciation for the remarkable progress in this field. If you have specific interests within LLMs, feel free to ask! I'd be happy to help you navigate this fascinating and ever-evolving landscape.