10 years of NLP history explained in 50 concepts | From Word2Vec, RNNs to GPT
From RNNs to Transformers to GPT-4, the leap in intelligence in Deep Learning research for Language Modelling and NLP has been a steady and educational growth. In this video, I explain 50 concepts that cover the basics of NLP like Tokenization and Word Embeddings, to seminal work like RNNs, Seq2Seq, Attention, to innovative Transformer models like BERT, GPT, XL-Net, and InstructGPT. I present the challenges we have faced in previous designs, and what the current architectures do to improve it, as well as highlight areas of improvement for future research.
For an overview of multimodal models that combine text with other input modalities like images, videos, audio, check out:
The video is divided broadly into 5 chapters, each adding up to 50 concepts that chronologically cover the advancements of NLP.
0:00 - Intro
0:47 - Basics of Language Modelling
1:44 - RNNs, Seq2Seq, Encoder-Decoders
6:18 - Understanding Transformers
8:26 - LLMs - BERT, GPT, XLNet, T5
12:54 - Human Alignment, ChatGPT, GPT4
16:20 - Outro
Thanks for watching!
#deeplearning #machinelearning #gpt #ai
Papers referenced:
Word2Vec:
GRUs:
LSTMS:
Seq2Seq:
Enc-Dec Attention:
Attention is All You Need:
GPT:
BERT:
Relative Position Embeddings:
Transformer-XL:
T5:
XL-Net:
Performer:
Linformer:
Longfromer:
LORA:
GPT-3:
InstructGPT:
GPT-4 Technical Report:
1 просмотр
658
179
2 месяца назад 00:04:16 2
Twenty One Pilots - Overcompensate (Official Video)
2 месяца назад 00:26:32 1
Trent Reznor & Atticus Ross (NIN) Break Down Their Most Iconic Tracks | GQ
2 месяца назад 00:57:30 3
БЕZZУБЫЙ КИSЛ0ТНИК И К0Р0ЛЕВИШНА В БИКИНИ @VolodymyrZolkin
2 месяца назад 00:40:20 1
High Priest Of Saturn “Sons Of Earth And Sky“ (Full Album) 2016 Heavy Psychedelic Rock
2 месяца назад 00:15:06 1
Never Changes - All Intros to Fallout 1 2 3 FNV 4 & 76