Part.1
“说人话”的艰难历程
ChatGPT“说话”的节奏表明,它是一个词元一个词元逐步生成答案的(图片来源[1])
神经网络模型生成一句话的过程示意图。
of program excessive been by was research rate not here of of other is men were against are show they the different the half the the in any were leaved
cat through shipping variety is made the aid emergency can the
cat for the book flip was generally decided to design of
cat at safety to contain the vicinity coupled between electric public
既然两个词元也不够好,那再长一些呢?毕竟AI领域有一个“信仰”:大力出奇迹!
Part.2
并不新鲜的“新鲜事”
Circulation revenue has increased by 5% in Finland. // Positive
Panostaja did not disclose the purchase price. // Neutral
Paying off the national debt will be extremely painful. // Negative
The company anticipated its operating profit to improve. // _____
OpenAI对于预训练得到的模型进行了新一轮的小规模训练,即有监督的微调(supervised fine-tuning,简称SFT)。他们招募了一个约40人的标注团队,对ChatGPT在不同场景下的输出进行了评估,看是否符合3H原则[6],并通过这种方式构建了一个含有约14,000语料的训练集。
3H原则
Part.3
都是概率惹的祸
看ChatGPT如何诠释“林黛玉倒拔垂杨柳”(图片来源[7])
Part.4
蛋白质是一种语言吗?
参考文献:
[1] The PyCoach. Using ChatGPT for Data Science.
https://medium.com/geekculture/using-chatgpt-for-data-science-ac5f8a00fb5a, 2022.
[2] Stephen Wolfram. What Is ChatGPT Doing … and Why Does It Work?
https://writings.stephenwolfram.com/2023/02/what-is-chatgpt-doing-and-why-does-it-work/, 2023.
[3] Shin, S. et al. On the Effect of Pretraining Corpora on In-context Learning by a Large-scale Language Model. In Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 5168–5186, 2022.
[4] Common Crawl. https://commoncrawl.org/.
[5] Sang Michael Xie and Sewon Min. How does in-context learning work? A framework for understanding the differences from traditional supervised learning.
https://ai.stanford.edu/blog/understanding-incontext/, 2022.
[6] Ouyang, L. et al. Training language models to follow instructions with human feedback. In Alice H. Oh and Alekh Agarwal and Danielle Belgrave and Kyunghyun Cho, editors, Proceedings of Advances in Neural Information Processing Systems, 2022.
[7] https://twitter.com/xwpajq/status/1629293200065036289, 2023.
[8] Madani, A. et al. Large language models generate functional protein sequences across diverse families. Nature Biotechnology,
https://doi.org/10.1038/s41587-022-01618-2, 2023.
[9] 深度剖析:ChatGPT 及其继任者会成为通用人工智能吗?| AI那厮.
https://mp.weixin.qq.com/s/j5xPMjrwTLconbUS4MTc9A