AI Exposed: Can Chat GPT Hold up in Research Science?

Подписаться 1,1 тыс.

50% 1

In this eye-opening video, delve into the reality of AI's role in scientific research with a fourth-year structural chemistry Ph.D. candidate. Discover the limitations and occasional failures of AI tools like ChatGPT when navigating the complexities of cutting-edge scientific inquiries. Join us as we test ChatGPT's capabilities and explore the crucial need for human oversight in the pursuit of scientific knowledge. Subscribe now for a thought-provoking exploration of AI's place in the world of research! Note that the research I conduct is typically secretive and I will not be showing examples relating directly to my own research.
_________________________________________________
Gear I Use
_________________________________________________
Cameras:
Sony FX3 Full-Frame Cinema Camera: amzn.to/42I7MlK
Sony A7RV Full-Frame Mirrorless Camera: amzn.to/3sm7Vi0
Sony 24 - 105 mm F4 G OSS Lens: amzn.to/42qqhLS
Sony FE 200-600 G OSS Lens (Renewed): amzn.to/44GCfBo
Sirui 50 mm T2.9 Full Frame Anamorphic: amzn.to/45kCXoX
Atomos Ninja V+ Monitor with ProRes RAW Recording and Extras: amzn.to/44gqlhL
Stabilization:
SIRUI SVT75 Pro: amzn.to/3NCijdr
SIRUI SVH15 Fluid Head: amzn.to/4audeNW
Ulanzi Zero F38 Carbon Fiber Travel Tripod: www.ulanzi.com...
Ulanzi E12 F38 Quick Release Fluid Video Head: www.ulanzi.com...
DJI RS3 Pro (amazon): amzn.to/3RAJvK3
DJI OM6 (amazon): amzn.to/3HZtfiH
Lighting:
Aputure 60XS (direct): shop.aputure.c...
Aputure 60 XS (amazon): amzn.to/49m2Qqk
Amaran 300c: shop.aputure.c...
Amaran Spotlight SE: shop.aputure.c...
Aputure Light Dome SE (direct): shop.aputure.c...
Aputure Light Dome SE (amazon): amzn.to/3SKmQxa
Calibrite Color Checker Passport for Video: amzn.to/3NLWDvS
GODOX TL30 K2: amzn.to/3ul7m5N
Audio:
DJI Mic: amzn.to/3DxTY3K
RODE Lavalier GO Professional Microphone: amzn.to/3PnCcpJ
SONY ECM-XM1: amzn.to/45DSMHC
Zoom F3 Field Recorder: amzn.to/3YKnWKF
_________________________________________________
I make commissions on purchases made through Amazon and Aputure links.

Опубликовано:

2 окт 2024

Ссылка:

Скачать:

Готовим ссылку...

Добавить в:

Мой плейлист

Посмотреть позже

Комментарии : 14

@chrisvsevil9790 7 месяцев назад

Good video, you definitely bring up some valid points, areas where AI is just not as effective as people would assume. Thanks again

@jackofmanytrades 7 месяцев назад

Thanks for the comment! Yeah, it’s definitely a tool with numerous limitations but in time, I’m sure things will improve a bit. I’m still skeptical that it’ll ever get to where it needs to be for research purposes.

@spelcheak 7 месяцев назад

It’s as bad as it will ever be again

@jackofmanytrades 7 месяцев назад

That’s true. Technology always improves.

@sanhej9870 7 месяцев назад

But why use v. 3.5 as an example? Gpt 3.5 was released 2 years ago and most llms are 10x better today

@sanhej9870 7 месяцев назад

Still a good video though :)

@jackofmanytrades 7 месяцев назад

Good point. I feel It’s probably something that most people have access to since some of the later versions are behind pay walls. In any case. My colleagues have been developing proprietary ML models specifically to tackle our problems and frequent mistakes occur. But the technology is definitely interesting. Thanks for the comment!

@ShA-ib1em 7 месяцев назад

@@jackofmanytrades Gbt 3.5 is relatively weak Gbt 4 is much better However your criticism is understandable for current AI ... But for future AI it doesn't stand. Gemini already has a mechanism to check against fabricated answers .. Data analysis requires specialized models.

@jackofmanytrades 7 месяцев назад

@ShA-ib1em the examples I shared are tip of the iceberg. I work on extremely complex systems which highly specialized AI models made by leading experts in my field cannot solve. AI poses some exciting opportunities but my point is that there are some problems particularly in science for which AI simply doesn’t have the ability to generate reasonable conclusions… especially when left unattended.

@ShA-ib1em 7 месяцев назад

@@jackofmanytrades I agree .. especially when you are on the cutting edge of science .. our models will struggle .. Because the edge is by definition a place where we don't have good data to train the models .. they need to train in order to learn to generalize .. However it's not "AI" in principle .. that is struggling .. it's our models or architectures .. that is still lacking .. but theoretically .. if we build better models .. in theory .. they can outsmart human reasoning..