Chapters (Powered by ChapterMe) - 00:00 - Introduction 00:17 - ChatGPT: Optimizing Language Models for Dialogue 00:39 - Examples 03:49 - Chat GPT and Instruct GPT 04:33 - Chat GPT - Learning 4 Stages 04:49 - Learn Word Solitaire 10:06 - The human teacher guides the direction of word solitaire 12:27 - Mimic human teacher preferences 14:13 - Learn from a simulated teacher with augmented learning 15:51 - Of course Chat GPT is still not perfect 17:17 - Epilogue - Socialization of Chat GPT
連續問了三個化學相關的問題, 其中第三個問題是"What background knowledge should I have before I start to learn about reaction mechanism?" 結果chatGTP想了很久然後就口吐紅字當掉. 前兩個問題回答得都不錯.
:Learning from the text doesn't guarantee the ability to answer questions properly. Is there additional tricks Open AI used? ChatGPT: Training a large language model on a large corpus of text data, as I was, does not guarantee that the model will be able to answer questions accurately or properly. This is because language is complex and unpredictable, and a model that is trained on text data alone may not have the necessary knowledge or understanding to answer questions accurately. To improve the ability of language models to answer questions accurately, OpenAI, and other researchers and developers, use a variety of techniques and strategies. These can include incorporating additional knowledge sources, such as encyclopedias, dictionaries, and other structured datasets, into the training process. This can provide the model with a more comprehensive understanding of the world and allow it to answer questions more accurately. Another technique that is commonly used is to fine-tune the model on specific tasks or domains. This involves using a smaller, task-specific dataset to fine-tune the model's weights and parameters, so that it is better suited to the requirements of the task. This can improve the model's performance on the specific task and allow it to answer questions more accurately. Overall, there are many techniques and strategies that OpenAI and other researchers use to improve the ability of language models to answer questions accurately. These can include incorporating additional knowledge sources into the training process and fine-tuning the model on specific tasks or domains. These strategies can help to improve the performance of language models and allow them to answer questions more accurately.
16:13 問沒用的問題 但我問它 Let W1 and W2 be subspaces of V. Prove that W1 U W2 is a subspace if W1 contains W2 or W2 contains W1. 它only if part回答是錯的捏... 還有問integrate((-1)^floor(1/x)/x) x from 1 to 2,它也亂回答 所以這些是沒用的問題嗎? 還是這是人類不想知道的問題? 😅
I have a mix feeling regarding to AI, but if I could have something, I would like to have chips that makes me speak 5 languages and obtain all kinds of knowledge like Lucy.