Large Language Model (LLM) ... Multimodal ... Foundation Models (FM) ... Generative Pre-trained ... Transformer ... GPT-4 ... GPT-5 ... Attention ... GAN ... BERT
Natural Language Processing (NLP) ... Generation (NLG) ... Classification (NLC) ... Understanding (NLU) ... Translation ... Summarization ... Sentiment ... Tools
Agents ... Robotic Process Automation ... Assistants ... Personal Companions ... Productivity ... Email ... Negotiation ... LangChain
SMART - Multi-Task Deep Neural Networks (MT-DNN)
Deep Distributed Q Network Partial Observability
Google
TaBERT
Google is improving 10 percent of searches by understanding language context - Say hello to BERT | Dieter Bohn - The Verge ...the old Google search algorithm treated that sentence as a “Bag-of-Words (BoW)”
Google AI’s ALBERT claims top spot in multiple NLP performance benchmarks | Khari Johnson - VentureBeat
RoBERTa:
- RoBERTa: A Robustly Optimized BERT Pretraining Approach | Y. Li, M. Ott, N. Goyal, J. Du, M. Joshi, D. Chen, O. Levy, M. Lewis, L. Zettlemoyer, and V. Stoyanov
- RoBERTa: A Robustly Optimized BERT Pretraining Approach | GitHub - iterates on BERT's pretraining procedure, including training the model longer, with bigger batches over more data; removing the next sentence prediction objective; training on longer sequences; and dynamically changing the masking pattern applied to the training data.
- Facebook AI’s RoBERTa improves Google’s BERT pretraining methods | Khari Johnson - VentureBeat
Google's BERT - built on ideas from ULMFiT, ELMo, and OpenAI
Attention Mechanism/Transformer Model
- Generative Pre-trained Transformer (GPT)2/3
Conversational AI ... ChatGPT | OpenAI ... Bing/Copilot | Microsoft ... Gemini | Google ... Claude | Anthropic ... Perplexity ... You ... phind ... Ernie | Baidu
- ChatGPT is everywhere. Here’s where it came from | Will Douglas Heaven - MIT Technology Review
Transformer-XL
Microsoft makes Google’s BERT NLP model better | Khari Johnson - VentureBeat
Watch me Build a Finance Startup | Siraj Raval
Smaller, faster, cheaper, lighter: Introducing DistilBERT, a distilled version of BERT | Victor Sanh - Medium
TinyBERT: Distilling BERT for Natural Language Understanding | X. Jiao, Y. Yin, L. Shang, X. Jiang, X. Chen, L. Li, F. Wang, and Q. Liu researchers at Huawei produces a model called TinyBERT that is 7.5 times smaller and nearly 10 times faster than the original. It also reaches nearly the same language understanding performance as the original.
Understanding BERT: Is it a Game Changer in NLP? | Bharat S Raj - Towards Data Science
Allen Institute for Artificial Intelligence, or AI2’s Aristo AI system finally passes an eighth-grade science test | Alan Boyle - GeekWire
7 Leading Language Models for NLP in 2020 | Mariya Yao - TOPBOTS
BERT Inner Workings | George Mihaila - TOPBOTS
Top BERT Applications You Should Know About | Tanya Malhotra - MarkTechPost ... PaLM
BERT 101 | Hugging Face

How BERT works using a fun MadLibs example. BERT is like a super smart language model that understands words and sentences really well. Imagine you're playing MadLibs, and you have a sentence with a missing word. Let's say the sentence is:

I went to the _______ to buy some delicious ice cream.

You need to fill in the blank with the right word to make the sentence make sense. BERT does something similar, but with much more complex sentences. Here's how it works:

Tokenization: BERT first breaks down the sentence into smaller pieces, called tokens. In our MadLibs example, each word is a token. So, "I," "went," "to," "the," "_______," "to," "buy," "some," "delicious," "ice," and "cream" are all tokens.

Word Representation: BERT gives each token a special code to represent it. Just like in MadLibs, you might have a list of possible words that could fit in the blank, and each word has a number next to it. BERT does something similar by giving each token a unique code that helps it understand what the word means.

Context Matters: BERT doesn't just look at one token at a time; it pays attention to all the tokens in the sentence. This is like looking at the whole MadLibs sentence to figure out what word fits best in the blank. BERT considers the words that come before and after the blank to understand the context.

Prediction: Now, the magic happens. BERT tries to predict what the missing word is by looking at the surrounding words and their codes. It's like guessing the right word for the MadLibs sentence based on the words you've already filled in.

Learning from Data: BERT got really good at this by studying lots and lots of sentences from the internet. It learned to understand language by seeing how words are used in different contexts. So, it's like you getting better at MadLibs by playing it over and over.

Results: BERT gives a list of possible words for the blank, along with how confident it is about each one. It can even give more than one option, just like you might have a few choices in MadLibs.