@timcarambat: ChatGPT is cool, but how did OpenAI get it to be so good? It was trained on gigabytes of text information posted all over the internet and trained on that. The set of data OpenAI used is private, but their is an open source equivalent that you can read! Could train your own language learning model now! #chatgpt #chatgpt3 #ai #aitools #gpt3 #gptt #programming #developer #learntocode #Tech #techtok #techtoktips #creator #openai #artificialintelligence

tcarambat
tcarambat
Open In TikTok:
Region: US
Wednesday 28 December 2022 20:39:03 GMT
35081
1638
96
150

Music

Download

Comments

fsgbrbfsg
Felipe :
GPT3 was trained on 44TB according to co founders. GPT2 was trained on 40GB.
2022-12-28 23:36:31
12
kamel3d
KAMAL :
40 gig? Thats not a lot
2022-12-28 23:26:25
10
smilespergallon
SmilesPerGallon :
Exactly, I put it to the test and exposed many biases.
2022-12-28 21:19:29
3
polymathicman
Jeff :
And how close to zero is the chances the copyright of the text owners are respected.
2022-12-28 21:57:07
1
ai_tinkers
AI tinkerer :
So much good information packed into this single tiktok! Nice! And ChatGPT is completely biased 😆😆😆
2022-12-29 01:20:28
7
davidco075
davidco075 :
now how to I train a model on this data?
2022-12-29 06:35:42
4
inssa_
Inssatoure :
Is it possible to have an working ai that can be used offline ?
2022-12-29 11:25:02
1
ltdata1
Lt Data :
Now if they can teach the ai to tell fact from fiction by looking at sources etc, I'd be very interesting to see some conclusions it comes to
2022-12-28 22:44:44
3
crftzman
crftzman :
it's from humans until it can independently connect the dots
2022-12-29 08:04:25
1
the_shah_of_blah
Testing 123 :
So the juice is in the model. Can you share some info on how these LLMs would work? Would like to get learn up a bit about them
2022-12-29 04:01:12
3
thestackhero
thestackhero :
Thanks for sharing
2022-12-30 01:42:24
1
martin_jam_
MartinJ :
I always knew the likes of Reddit and Quora would be used for AI.
2023-01-01 07:27:13
1
nashadelic
Nash :
Thanks for doing this!
2022-12-29 19:28:05
1
mrfymrf_3d
MrfyMrf_3D :
I'm guessing this is what Cliff High has been talking about for years with his 'web bot'..? Great video pal
2022-12-29 08:40:10
1
theshawwn
Shawn Presser :
I’m one of the authors of The Pile. If you haven’t heard of it, you may want to check it out. It’s an 800GB dataset. I scraped 19000 books (“books3”).
2023-01-02 06:08:31
3
tropaion
tropaion :
I thought the used "The Pile"
2022-12-30 00:06:41
2
kamel3d
KAMAL :
So they have lot of random text but how would they know what is correct and what is wrong and based on what?
2022-12-28 23:29:44
1
kamel3d
KAMAL :
Next show us how to train your own chatgpt 😂
2022-12-28 23:30:38
1
sensuous_lifestyle
Sensuous Lifestyle :
Good point.
2022-12-30 05:49:27
1
anonymous____________12
Anonymous____________12 :
chat gpt is a left leaning, global warming consensus nonsense defender😳
2022-12-29 06:40:21
1
killswitchhbk
KillswitchHBK :
That’s cool 😵‍💫😵‍💫😵‍💫
2025-04-08 13:08:47
1
antonio.sergioii
Antonio Sérgio :
So, now can we use online GPUs services and those presets in order to build our own chatGPT?
2023-01-04 04:49:52
0
mustafamsy
Mustafa Alyousef :
Imagine what Elon will do with Twitter when Reddit was used for data to train the AI
2023-01-04 20:39:20
0
thesocialsniper
thesocialsniper :
Could you show a practical example on how to feed new data into GPTJ?
2023-01-03 22:00:52
0
engineer_2_educator
Engineer_2_Educator :
Thanks
2022-12-28 23:43:34
0
To see more videos from user @timcarambat, please go to the Tikwm homepage.

Other Videos


About