In a paper, researchers test the best language models (GPT-2/3/Neo) at solving programming questions from coding interviews. Results aren't particularly groundbreaking but show potential.
Insider preview boosts ransomware defence, fixes memory leaks, more
windows 10 21h2
Build 22000.282 fixes for Ryzen L3 cache performance issue, and more
windows 11 insider preview
Transformer models RSS
DALL·E, a distilled version of GPT-3 with 12 billion parameters, is a transformer model that produces images from a given caption. The results are impressive with the model capturing minute details.
OpenAI has exclusively licensed the largest transformer model to date—GPT-3—to Microsoft. The latter will use GPT-3's NLG and NLP capabilities in building AI solutions for its customers.
Language-agnostic BERT Sentence Embedding (LaBSE), is a multilingual BERT embedding model that supports language-agnostic cross-lingual sentence embeddings for 109 languages by combining MLM and TLM.