logseq/pages/hls__Ozkaya - 2023 - Application of Large Language Models to Software E.md

file:: [Ozkaya - 2023 - Application of Large Language Models to Software E.pdf](file://C:/Users/david/Zotero/storage/WHX3PLDV/Ozkaya - 2023 - Application of Large Language Models to Software E.pdf) file-path:: file://C:/Users/david/Zotero/storage/WHX3PLDV/Ozkaya - 2023 - Application of Large Language Models to Software E.pdf

• not to mention software engineers ls-type:: annotation hl-page:: 1 hl-color:: green id:: 65082008-a809-498c-aafb-913403fc7463
• Recently released LLMs, such as Generative Pretrained Transformer(GTP) 4 used in ChatGPT by OpenAI and BERT used in Bard by Google, disrupt the search engine model that we have been used to. ls-type:: annotation hl-page:: 1 hl-color:: blue id:: 65082aeb-6c17-47ac-b388-90ea6f5c361a
• An LLM is a deep neural network model which has been trained on large amounts of data, such as books, code, articles, and websites, to learn the underlying patterns and relationships in the language that it was trained for. ls-type:: annotation hl-page:: 1 hl-color:: purple id:: 65082b88-cec6-4393-af1d-3f5b933fac38
• the model is able to generate coherent content such as grammatically correct sentences and paragraphs that mimic human language or syntactically correct code snippets ls-type:: annotation hl-page:: 1 hl-color:: green id:: 65199d20-c3da-4dac-89e5-b51167b719b2
• While the content generated by LLMs are often grammatically correct, they may not always be semantically correct. ls-type:: annotation hl-page:: 2 hl-color:: blue id:: 65199d44-84bc-4ed0-924e-874a4f8173be
• The probabilistic and randomized selection of the “next token” in constructing the outputs on one hand gives the end user the impressions of correctness and style, on the other hand may result in mistakes. ls-type:: annotation hl-page:: 2 hl-color:: green id:: 65199da0-f6b8-495d-a04d-07f8bef810e2
• Data quality and bias concerns: ls-type:: annotation hl-page:: 2 hl-color:: blue id:: 65199dc4-a368-4507-b421-6cc3f772f621
• Any of the issues that exist in the training data, such as biases and mistakes, will be amplified by LLMs ls-type:: annotation hl-page:: 2 hl-color:: green id:: 65199ed1-141f-4836-b26f-2da8c1687791
• making prejudiced recommendations ls-type:: annotation hl-page:: 2 hl-color:: purple id:: 65199ee0-a186-43d2-a8df-7d3eabbb7498